Innovation Meets Reality: The SKA Design

Abstract

This chapter is an account of the creativity of a host of engineers and scientists to build a groundbreaking radio telescope with the performance needed to extract the scientific treasures buried in weak radio signals from the distant Universe, of which there were previously only hints. Always moving forward, it describes the road from the ambition to overcome barriers and harness new technologies to the reality of thousands of difficult individual decisions that had to be made to successfully yield a realisable design in the available time. The global milieu in which this was done as an international project has set many precedents.

This book describes the roles of many factors: persistence, management, luck, competition, cooperation, and politics. The way the SKA came together represents a global first among international science projects, even in the time horizon of this book, and in the end will indeed result in two next-generation radio telescopes. This chapter traces the evolution of the roles of innovation, technical development and engineering approaches, alongside the other chapters that describe the many other aspects required to make the SKA happen. The story of the winnowing of technology provides a lesson that it is not easy to accomplish the twin goals of changing established technology at the same time as successfully funding and building a major new facility.

The originators of the SKA idea in the 1990s understood that a next-generation radio telescope was needed, but also realised that breakthroughs in technology or techniques would also be needed to achieve their vision. The SKA began with a serious attempt to transform the design of radio telescopes to enable a next generation that would surpass the performance of the then current generation by 100 times at an affordable cost. Although sounding extremely ambitious, this was known to be possible in other fields, such as particle physics and the miniaturisation of electronics. These aspirations triggered a period of invention, adaptation, and innovation, and inspired a global pool of hundreds of talented engineers and scientists. It also attracted sponsorship by governments and funding agencies.

Innovation, used to capture all the above, takes many forms and is difficult to capture precisely. It is usually considered very disruptive if it completely obviates current methods or existing technology, but this is rare. Most innovations are mildly disruptive, and some are just incremental improvements on existing technology. For example, while the invention of the laser was much more than mere innovation, it provided opportunities for thousands of innovations which emerged. The most important innovations incorporate the concept of a platform, a framework which supports many capabilities. As will be illustrated, this concept applies to general-purpose radio telescopes too, in the sense of enhancing the likelihood of new discoveries, which by definition cannot be specifically designed for. Almost all the new approaches to the design of the SKA were directed to enhancing discovery space.

Innovation cannot be predicted or planned for, but circumstances can be created in which its probability is enhanced. In the case of the SKA in the late-1990s and early-2000s, the circumstances were ripe for new approaches to radio telescope design.

Initially, innovations tend not to work as well as existing approaches or designs. This is a natural barrier, which can delay or even halt their usage. It is conceivable that some of the many new ideas that were explored for the SKA were halted prematurely and some might even become practical in the future if additional developments occur. This is because an innovation period can last only so long before some retrenchment is in order. As described in Chaps. 3 and 4, the need for a proven design became more pressing as funding agencies became interested. At that point the SKA had to make a transition from focussing on innovative designs to the engineering required to deliver a realistic next-generation telescope. However, the design path was not straightforward, and this chapter tells the story of how it occurred.

Although every project is unique, it is hoped that this chapter might also shed light on age-old themes that seem to pervade highly technical projects, as they have manifested in the SKA project. Cross references to the following points are made throughout the chapter.

• (A) The role of optimism. Where is the line between boundless optimism and that which is necessary to propel innovation? A famous quotation from Robert Noyce frames the issue¹: “Optimism is an essential ingredient of innovation. How else can the individual welcome change over security, adventure over staying in safe places?” A balance must be struck, because optimism to the point of not recognising challenges will eventually lead to poor results. Optimism played a definite role in driving the SKA project forward.

• (B) The role of discovery space in the design of large scientific facilities. In radio astronomy almost all major discoveries have been unexpected outcomes of observations with telescopes not designed to make such observations (e.g., Fast Radio Bursts (Lorimer et al., 2007)) (see Sects. 1.2.2 and 5.3.7). Widening parameter space, increasing flexibility, and promoting agility in the final design all enable larger discovery space. But these design approaches are usually expensive, and their usefulness cannot be predicted in advance. In contrast, a goal-oriented design sets out to answer a specific scientific question. Threading the needle between these extremes created a consistent underlying tension in the years leading up to a stable design between maintaining a high degree of flexibility and achieving a practical, affordable design.

• (C) The psychology of cost and cost projections. In a subject related to the previous two, realistic projections of cost and risk, if made too early, are likely to kill a project in the eyes of all but the most sophisticated national funding agencies. Sensitivity to costs was intensified by the need for funds to build large prototypes to fully test SKA antenna concepts. It is conceivable that some SKA concepts might have borne fruit if sufficient funds and time were available to develop them to maturity, which would then enable accurate cost estimates (see Sects. 6.4.4.2 and 6.4.6). The progression of cost and schedule targets for the SKA project is discussed in Sect. 4.6.1.

• (D) Born Global. SKA started as a global project without a single major sponsor providing most of the resources and driving decisions. This is unusual in astronomy. What specific lessons can be learned from the SKA project, especially in its formative years? Projects as large as the SKA tend to force international collaboration because most nations cannot afford the cost. But international collaboration almost always costs more than one carried out efficiently under one sponsor, and as described in Chaps. 1, 4 and 11, it takes much longer. Also, although technical risk is likely to increase because there are ‘too many cooks’, the risk of cancellation is likely to decrease, because the risk of one sponsor withdrawing is diluted (see Chaps. 4 and 11).

• (E) Complex Project. As described in a series of workshops by Gary Sanders (California Institute of Technology),² most global projects exhibit the characteristics of complex projects: multiple resource bases, political interference, clashes of institutional cultures, and several others. National interests have the potential to trump everything in international complex projects, sometimes to the detriment of lowest cost or best performance possible. Constant independent scrutiny is clearly needed. There are likely to be many similar complex projects in the future. Will the SKA survive complexity? The SKA project has clearly survived the challenges of complexity up to the time of writing, including during the period covered in this book (see Chaps. 1 and 11).

• (F) Non-science benefits. Excellent science is a necessary condition for institutional funding and support, but governments and funding agencies also value benefits to industry, employment, the potential for innovation and access for their scientists. Unless there are identifiable economic benefits, prospects for support are diminished. While involving industry directly is mainly beneficial, issues with intellectual property can also hinder progress by impeding essential dissemination of information.

• (G) Evolution of technology. As discussed in Sect. 1.3, Livingston curves (Livingston, 1954), log-plots based on the evolution of particle accelerator energy, were used by analogy to illustrate the evolution of radio telescope sensitivity in the formative years of the SKA and to promote innovation and the project in general. Moore’s law (Moore, 1965) and the growth of optical fibre communication, resulting from the invention of the laser, are examples of exponential growth as well (see Chap. 1 and SKASUP1-1). Underlying inventions or discoveries are required to set the scene for these developments. Even if underlying origins are present, mature technology often presents a barrier to innovation that must either be overcome or tunnelled through (See Sect. 11.4.3).

• (H) Helpful technology diffusion. Technology developments occurred after 2000 that enabled the SKA to reach its current construction stage (e.g., optical fibre data transmission). Further developments are likely to be needed to proceed to the originally envisioned SKA Phase 2. Would any of the innovations investigated in the SKA’s formative stages survive if they were re-started today?

• (I) Unhelpful technology diffusion. The accelerating utilisation of the radio spectrum for other uses may gradually choke off its use for astronomy. More recently, the capability to launch thousands of small satellites for the first time has resulted radio frequency interference directly in the look-direction of radio telescopes. Weak spectral lines and highly redshifted spectral lines will be most at risk. Will the SKA be the last major ground-based telescope to access a scientifically useful fraction of the radio spectrum for astronomy? (See Sect. 6.2.2.12).

• (J) The need to deliver. Once it has become clear that there is limited time available to fully develop more risky innovations, large project management and system engineering processes must take hold. This discipline will generally slow down change in the project design and architecture, sometimes to the detriment of science goals. Did this happen at the right time for the SKA? (See Chap. 11).

• (K) Project Management and System Engineering. Do the disciplines of project management and system engineering also lead to a form of tunnel vision, especially projects that take more than a decade to build? This is their intent. Otherwise, the design will never converge, and there is no other method of managing a large-scale project involving many parties and people. Achieving a balance between creative change and accountable management is a complex problem.

6.1 Design Goals for a General-Purpose Radio Telescope

The SKA has always been considered a next-generation general-purpose telescope for all of metre and centimetre wave astronomy. This role is arguably held now by the Jansky Very Large Array (JVLA) in the USA, which was commissioned in 1980 under the name, Very Large Array (VLA), and upgraded over a period from 2001 to 2012 (Perley et al., 2009, 2011). Although scientific goals were elucidated in its funding proposals, it is a general-purpose telescope, not designed to carry out specific science experiments. Its capabilities are broad, and its record of scientific achievement is amazing.

The key aspects of the upgrade in the early 2000s were continuous frequency coverage (1–50 GHz), increased sensitivity, increased instantaneous bandwidth and higher frequency resolution. The upgrade provided greater access to new discovery space, primarily new frequencies.

The design of the SKA follows a similar general-purpose philosophy. Its most important design aspects incorporate the concept of a platform, i.e., a framework which supports many capabilities. The approach is to produce a constrained maximisation of discovery space by retaining flexibility wherever possible. Flexibility is abandoned only when it strikes limitations of technical feasibility and cost at the time of its design. As will be described in this chapter, many innovations were explored in depth, all of which were intended either to reduce cost so that “more telescope” could be afforded or to directly increase discovery space.

For a radio telescope, one can describe the capabilities of the platform as a multi-dimensional parameter space, each of which confers a significant capability, but to describe this space in detail is beyond the scope of this book. In terms of capabilities, the important parameters enable the telescope to make 3-D radio images (two dimensions on the sky and one frequency dimension) with high sensitivity and resolution, so that narrow spectral lines can be traced in the frequency dimension. Moreover, the telescope must be able to capture all the spatial scales in complex images while at the same time covering large regions of sky. Because much of the sky contains polarised radio emission, it is also important to be able to measure polarised radio emission in the images. Finally, there are time-variable sources of radio emission whose time scales vary from milliseconds to years.³ These capabilities were all driven from what was perceived as the most important science questions in astronomy (see Chap. 5), while recognising that historically, the most important discoveries have been unexpected. In such cases a telescope happened to detect a phenomenon for which it was not designed. Chapter 11 discusses the recent example of the CHIME telescope, which has found most of the phenomena of Fast Radio Bursts, despite being designed for something completely different.

To span the required range of frequencies with this set of capabilities required two separate telescopes now under construction in Australia and South Africa. By 2012, the end of the period covered in this book, the design had incorporated design improvements in almost all the capabilities described above, which are not available in today’s telescopes.

6.2 SKA Innovation History

The history of major telescope developments from the 1950s to 1990 is covered in detail in Chap. 2. This history of this period illustrates the mind-sets of radio astronomers which converged from several independent threads of thought circulating in the global community. Interferometry, utilised as arrays of antennas in various forms, could provide high resolution. But because radio astronomy signals are very weak, especially the ubiquitous hydrogen line at 1420 MHz, radio telescopes also needed high sensitivity to progress the science. Sensitivity, especially at relatively high radio frequencies, was difficult to obtain at reasonable cost. More subtly, obtaining good radio imaging requires a continuous representation of interferometer spacings in array designs. This thinking was on display in 1990 at the IAU Colloquium 131 on Radio-Interferometry (Cornwell & Perley, 1990) but it took until 1993 for it to consolidate (see Sects. 2.4.3–2.5). This chapter briefly covers events of the 1990s, with increasing detail from 2000 to 2007, yet more detail from 2008 to 2012, and in some cases extending a bit further. As at the time of writing the SKA is under construction, there will be much more to tell in the future.

Figure 6.1 is a guide to the technical history of the project and provides a link to other chapters in this book (see also Figs. 3.1 and 4.1). Exploratory technical discussions took place from the early 1990s to 2002, during which major prototypes were built and performance tests carried out. Table 6.1 is a companion to Fig. 6.1, and contains a table of important documents along with notes as to their significance.

Fig. 6.1

A timeline of the technical development of the SKA (Acronyms and Abbreviations. PreCon: Pre-construction, LTWG: Large Telescope Working Group, URSI: International Union of Radio Scientists, MoA: Memorandum of Agreement, ISSC: International SKA Steering Committee, SSEC: SKA Science and Engineering Committee, ISPO: International SKA Project Office, SPDO: SKA Program Development Office, PrepSKA: Preparatory phase for the Square Kilometre Array, SKADS: SKA Design Study, AAVP: Aperture Array Verification Program, TDP: Technology Development Project (USA), 1st Conv. Rep.: First Convergence Report, Exp. Astr.: Experimental Astronomy, ESFRI: European Strategy Forum on Research Infrastructures, Options T.T.: Options Tiger Team, Sys. CoDR: System Concept Design Review, ASTRONET: network of European funding organisations, ASTRO2010: the 2010 report from the US Decadal astronomy survey in 2010, PEP: Project Execution Plan). The top part illustrates the context of governance and funding (see Chaps. 3 and 4). The bottom part shows the dates of publication of SKA Memos and other papers that were influential in subsequent technical developments. The red-coloured items denote major turning-point documents. The Mxx notation stands for SKA Memo xx

Table 6.1 Documents representing major turning-points in SKA development

6.2.1 1993–2007: The Pre-PrepSKA Period

6.2.1.1 Beginnings of Engineering Coordination

Initiatives to develop new technology and engineering approaches to the design of the SKA had been taking place in the most prominent institutes around the world since 1993, cooperating and communicating through the URSI Large Telescope Working Group (LTWG) (see Sect. 3.2.1). More formal cooperation was put in place in 1996 (see Sect. 3.2.2) with a Memorandum of Agreement which was signed by eight globally- distributed⁴ institutions to cooperate in a technology study program leading to a future very large radio telescope.⁵ As described in detail in Chap. 3, the International SKA Steering Committee (ISSC) was established in 1999 and met approximately every 6 months, ending in 2007. A key event occurred on January 1, 2003, the establishment of the International SKA Project Office (ISPO) with Richard Schilizzi as its director.

From a technical perspective, the ISSC set itself a goal of increasing telescope sensitivity by a factor of 100. Not only did subject matter experts believe that this ambitious goal could be achieved, they also each had in mind innovations to achieve it. But only after the establishment of the ISSC did serious critical examination of the various proposals begin to take place.

The goal was solidified in a synopsis of technical specifications assembled by Ron Ekers⁶ in late-2001, based on science and engineering discussions in preceding years. These specifications, rooted in science, provided more specific engineering direction.

The meeting of the ISSC in August 2000 recognised the need to coordinate engineering effort. Based on a recommendation from the Five-year Management Plan and Technical Oversight Working Group, an Engineering Management Team (EMT) was established (later renamed International Engineering Management Team—IEMT). Their remit was all-encompassing: produce a technical audit of SKA technical activities, identify deficiencies, maintain an evolving SKA system definition document, foster information flow among project groups, and coordinate with a similar-level Science Advisory Committee on interacting issues. After some iterations, a group of interested but independent people were appointed.⁷ Peter Hall (Australia Telescope National Facility) was appointed by the ISSC as chair. From then until about 2007, Hall dominated the leadership of the international SKA engineering scene.

Apart from the SKA Newsletter series, itself, this period is also covered by notes made from the newsletters,⁸ which contains lists of events and technical meetings, as well as notes ordered by SKA-related entities (countries and EU-related organisations).

6.2.1.2 2003–2007, the International SKA Project Office (ISPO): Working Towards Engineering Coherence

In 2003, Schilizzi put forward an SKA Management Plan,⁹ which formed the basis for management of the ISPO. This plan re-enforced the role of the IEMT as a key part of SKA Management but changed the emphasis to “conduct reviews of national and regional design studies” and “act in the capacity of (an) engineering working group”. Table 6.2 contains a list of innovative receptor-concepts¹⁰ being put forward by national groups. The role of the IEMT in the management plan was a step towards a tangible technical design for the SKA, while retaining the activity of winnowing options. In 2004, the IEMT was renamed the Engineering Working Group (EWG) .

Table 6.2 SKA concepts in the 2001–2005 period

Table 6.3 shows a trail of documents in which the EMT/IEMT carried out their mission to review the concepts in Table 6.2. After putting out a request for technical information,¹¹ the group produced the first comprehensive assessments of the design studies (SKA Memo 27).

Table 6.3 Major engineering documents in the pre-PrepSKA period

One aspect of their assessment stands out: “The EMT is persuaded that independent, widely-placeable, multi-beams translate into a true sensitivity gain for a radio telescope.” This was widely accepted at the time, but it is interesting to note that only at low frequencies has this been realised in the SKA design (i.e., wavelengths longer than about 1 m). The reasons are discussed in Sects. 6.4.5 and 6.5. Independently, John Bunton (Australia Telescope National Facility) attempted to evaluate the designs based on cost as a function of frequency,¹² which turned out to require many assumptions and succeeded only in highlighting some rough trends.

The assessments of late 2003, documented in SKA Memo 41 (see Table 6.3), carried out by working groups appointed by the IEMT, were impressively thorough. A key aspect was “hybrids” or “composite” solutions involving a combination of concepts. While noting that “No one concept provides optimal performance in both the high frequency and multi-fielding domains”, they also cautioned that “The IEMT feels that investigation of hybrids, while important, should not de-focus efforts in key development areas within each concept”. The obvious hybrid was a frequency-split between two technologies. However, almost all technologies claimed high sensitivity from about 150 MHz to 10 GHz. Although work continued apace on most of these concepts, Luneburg Lenses¹³ were dropped in 2004 after a review by the Australian SKA Consortium.

Based on the SKA meeting in Penticton, Canada in July 2004, a book was published in 2005 by Springer, edited by Hall (Hall, 2004a), who was now SKA Project Engineer. This volume contained a compendium of aspects of all the SKA demonstrator projects and many other SKA technical developments. Hall’s overview paper (Hall, 2004b), summarising the status at time, pointed out that many specialised technologies needed to be refined, not just antennas. One concern, for example, bubbling below the surface, was that the cost of image processing¹⁴ (see Sects. 6.6.5, 6.2.1.3, and 6.4.3.1) would scale as the inverse 8th power of antenna diameter.

The way forward was now seen as “convergence”, another way of expressing the idea of a hybrid solution. The ISPO organised a workshop in South Africa in January 2004, explicitly set up to discuss options for hybrid designs and to “explore the parameter space of combined designs and narrow down the possibilities to a small number that can be focussed on in the future”. Schilizzi’s summary¹⁵ re-enforced the emerging themes noted above, while stressing that cost must be contained to €1 billion and that “a mutually agreed single concept that is inclusive and engages the global community” was required. Several types of hybrids were identified, including a “site hybrid”, which turned out in the end to be what took place.

The summary contained an important table illustrating the depth of expertise in the six countries participating at the time. For each of the 21 technical areas, there were at least three countries where significant expertise was present. In essence, this was a global dream team, which could capably explore all the innovative approaches if provided with appropriate resources. In the event, however, they never operated as a global team. Although there were a few exceptions, there was not much cross-fertilisation of ideas. Each country concentrated mainly on developing their own concepts, rather than supporting concepts elsewhere. Nevertheless, the IEMT did provide excellent feedback on the strengths and weaknesses of each concept, but that is a long way from actual participation. This is clearly one of the features of a complex project (see point {E} in Chap. 6 the introduction to this chapter).

In the meantime, SKA specifications were firming up and becoming more detailed.¹⁶ However, sensing that primary goals might be lost, scientists stressed the need at the South Africa meeting to maintain sensitivity as the most important parameter, more so than complete frequency coverage, and the multi-fielding needed further investigation for scientific benefits.

Six months later, the initial convergence meeting was followed by another in which more detailed discussions and analyses of specific concepts and technical risks took place. The second summary¹⁷ contains more detail, complete with the meeting presentations. Of particular importance was a presentation by Bruce Veidt (National Research Council of Canada, Dominion Radio Astrophysical Observatory), which contained a thumbnail analysis of each potential hybrid, and for the first time discussed the issue of technical risk¹⁸ (see also Table 6.3). However, against the flow, Ken Kellermann (US National Radio Astronomy Observatory) expressed concern in a presentation at the meeting that hybrids would simply increase cost without yielding significant science benefits and was not sympathetic to keeping a broad spectrum of participants involved.¹⁹

These meetings were a turning point in the life of the SKA. Pressure was building to come to agreement on an SKA design. The winnowing process had begun in earnest, even though it was still below the surface.

In parallel with the engineering management activity, Hall also pioneered policy on the SKA relationship to industry.^20,21 This was important groundwork for later industry interactions (see Chap. 10). This work had political implications because the sponsors (governments) were keenly interested in economic benefits. The SKA presented many opportunities for such benefits, but also led to conflicting concerns over intellectual property (IP) (see point {F} in Chap. 6 introduction).

The EWG (renamed from the IEMT) continued IEMT practice, carrying out annual evaluations of the SKA demonstrator projects in 2004,²² 2005²³ (Table 6.3) and 2006.²⁴ All were given a numerical score, based on a large list of criteria that included cost, risk, schedule, security of funding, project management and responsiveness to questions. There were no stellar results. It continued to be obvious that many engineering challenges lay ahead.

A compendium of the work of the EWG (including the earlier work of the IEMT) through its task forces was published in 2007.²⁵

6.2.1.3 The Reference Design

As described in Sect. 3.4.1, the Heathrow meeting in June 2005 was a turning point for the SKA because it was now ‘on the radar’ of the funding agencies. But as the SKA project had equally weighted several different technical concepts, the funding agencies were concerned that the SKA was too immature to provide near-term support.

Possibly the lack of high marks for any of the alternatives by the EWG (see the previous section) and other indicators also led the ISSC to realise that the project was at an impasse. This was a fair assessment. Discussions in ISSC meetings subsequently led to the definition of a Reference Design for the SKA.²⁶ And following discussion at the November 2005 meeting, the ISSC instructed the Director to put together a ‘tiger team’ of ten prominent people to provide the Reference Design.²⁷ Their approach was that the design “should contain a substantial component of known technology in order to minimise risk yet should include an innovative component that enables access to new scientific parameter space, is challenging for the engineering community, and is attractive to policy makers and industry”.

For mid-frequencies, the reference design report notes that dish-based technologies are inherently more mature than aperture array technologies. Indeed, although not the final step, this was a dramatic step in the direction of abandoning some of the innovative ideas of the previous decade in favour of a buildable interferometric telescope within 5 years (i.e., lowish risk). This trend is traceable to the issues expressed more generally in the introduction to this chapter (see points {G}, {J} in Chap. 6 introduction).

The key features of the reference design were:

• Frequency coverage from 100 MHz to 25 GHz.

• An array configuration with baselines up to 3000 km, with half the collecting area in a central core of about 5 km in diameter.

• Small dishes (in contrast to Large Diameter) with “smart feeds”. These were very small dishes (about 10 m).

• Aperture Array tiles in the core with multiple independent fields-of-view, covering 0.3–1 GHz.

• An ‘Epoch of Reionisation’²⁸ (EoR) array also in the core area, like the LOFAR design.

• Supporting technologies such as data transport, processing, and software to be common to all forms of collecting area.

Recognising the nature of telescope arrays, a phased approach, with the first phase having about 10% of the total collecting area, was described in some detail. All the technologies were represented in the Reference Design apart from the EoR array, which was thought to be unnecessary, since a 10% version would have been no larger than LOFAR was planned to be. However, by 2007 this view had changed, when an EoR array was included in the SKA Phase 1 concept²⁹ (see Sect. 5.9.5).

Although the Large-N–Small-D (LNSD)^30,31 approach was a standard radio telescope design, the aim of the Reference Design was to avoid choking off innovation that might still be possible within the Reference Design framework, perforce deliberately entailing a large measure of risk. For example, the term “smart feeds” meant a combination of Phased Array Feeds (PAFs) (Sect. 6.4.7) and Wide-Band Single Pixel feeds (WBSPF) (Sects. 6.4.5 and 6.6.1.1). Both were considered “known technology”. Although not as definitively, Dense Aperture Arrays (Sect. 6.5.5) were considered quite mature. Unfortunately, none of the projects, put forward over-optimistically as mature, managed a sufficiently convincing level of success to be incorporated into the final SKA design,³² one of several instances of over-optimism (see point {A} in Chap. 6 introduction).

The selection of the LNSD approach to the SKA eliminated further consideration of the FAST (initially called Kilometre-Square Area Radio Synthesis Telescope (KARST)³³) and LAR designs (see Table 6.2) for the SKA. The early background to FAST is described in Sects. 3.2.6.2, 3.3.3.3 and the China’s proposal for the SKA site, utilising FAST-like antennas, in Sect. 7.3.5. A brief outline of the KARST/FAST design is described in SKASUP6-29.³⁴ The LAR approach is described in SKASUP6-28.³⁵

The US Technology Development Program (TDP) , funded in October 2007, formed an Antenna Working Group (AWG) to coordinate dish development (see Sect. 6.4.5). Although their funding proposal³⁶ was broader, the main emphasis was to pursue the LNSD design philosophy.

The 10-m dish diameter in the Reference Design was a compromise. US participants favoured 6.5-m dishes like those of the in the Allen Telescope Array (ATA), based on the idea that small dishes inherently have wide fields-of-view and high survey speed, a key science goal (Sect. 6.4.3.1). On the other hand, there was concern for the computing costs of wide-field imaging. Computing costs were thought to scale as d⁻⁸, where d is the dish diameter.³⁷ Nevertheless, it did not prevent proposals for even smaller dishes, but only if intermediate beamforming were used to mitigate the computing cost³⁸ (Sect. 6.6.5).

6.2.1.4 Goal Posts for the Future: Memo 100

Between the dissemination of the Reference Design in January 2006 and March 2007, the ISSC had assembled an ‘options tiger team’, chaired by Schilizzi, to deliver an options report,³⁹ building on the Reference Design, in which many design aspects were deliberately left open. The team identified five options, combinations of technologies to cover fundamental capabilities, such as frequency range, sensitivity, field-of-view (FoV), and transients. Despite stating that “FoV expansion technology (aperture arrays (AAs), PAFs, and multi-cluster feeds) … will be incorporated in the SKA design if and when it proves feasible”, all the options contained these technologies, although AAs were included only as a transient monitor. It was also stated that no additional R&D was needed for the 100–300 MHz range, because this would be developed in the context of LOFAR, MWA or the LWA. Finally, it was noted that if dish cost or technology limited frequency coverage to less than 25 GHz, a second dish array would be required.

Phasing the telescope construction was strongly emphasised, with the proviso stated in the options report, SKA Phase 1 “must perform as a science instrument in its own right and form a coherent steppingstone to the full SKA”. A timeline was included in the options report: SKA Phase 1 was to begin construction in 2012 and to take about 3 years. In March 2007, the ISSC formally adopted a resolution endorsing a phased implementation.⁴⁰

As the phased approach was formally agreed and in light of the upcoming European ASTRONET Roadmap and US Decadal Review (see Fig. 6.1 and Table 6.1), it was decided that a formal set of revised specifications was needed, leading to a baseline implementation. A ‘specifications tiger team’⁴¹ was assembled to provide these and a clearly defined route forward through SKA parameter space. The outcome was SKA Memo 100, “Preliminary Specifications for the Square Kilometre Array”.⁴²

Cost issues were much more prominent now. Indeed, some of the innovations put forward were directed at reducing the cost of collecting area (e.g., mould-based reflectors (Sect. 6.4.4.2) and Preloaded Parabolic Reflectors (Sect. 6.4.4.3)), but these needed significant up-front funding and development time before costs could be estimated with reasonable accuracy (see point {C} in Chap. 6 introduction). The tiger team was cognisant of opposing considerations: nailing down SKA specifications and cost, while not dampening the optimism in the engineering community needed to continue working on the various concepts (see point {A} in Chap. 6 introduction).

Memo 100 was such a major step that it had to be independently reviewed. The SKA Specifications Review Committee (SSRC), chaired by Roy Booth (Onsala Space Observatory, Chalmers University of Technology), was formed, and forthwith issued its report.⁴³ Although the SSRC outlined many challenges to be met, PAFs and AAs were considered sufficiently mature and important enablers of high survey speed. A major recommendation was “… we suggest that the SKA project should have stronger interfaces with the pathfinder projects, and that many uncertainties may be alleviated through the actual use of the pathfinders like LOFAR and the EVLA.”.

A record of tiger team discussions⁴⁴ on the SSRC report provides a more concise view of their recommendations than the report, itself, and provides additional insight into the collective thinking about future directions for the SKA. The SPDO, the US Technology Development Program (TDP) and others emphasised that they were already linking their activities with the pathfinder projects. However, in subsequent years the record was quite spotty in this regard for many different reasons: political, personalities and competition for resources (see point {E} in Chap. 6 introduction) (see also Sect. 4.5.2).

Without a doubt, Memo 100 is one of the most important documents in the history of SKA development. It summarised top-level science goals, laid out key technical specifications required to achieve each category of science, covered potential technologies available to enable the specifications, considered cost implications, and outlined the construction phases and anticipated a construction timeline. This was the first time that all of this was available in one place, a detailed elucidation of the SKA dream. Although many of the details turned out to be far too optimistic, the science goals and the general approach remain as the SKA is being constructed.

6.2.1.5 The SKA Precursors: ASKAP, MeerKAT, MWA

The Australian SKA Pathfinder (ASKAP, see Fig. 4.2 and Fig. 6.10) began gradually with early interest in phased array feeds (PAFs) for dishes. The story of PAFs is told from various perspectives in this book: Sect. 6.2.2.5 for the impact on SKA development, Sect. 6.4.4.1 for antenna design and Sect. 4.3.3.1 for ASKAP, itself.

MeerKAT (see Fig. 4.4) also began gradually, with early experiments on dish fabrication from composite materials, which is described in Sect. 6.4.4.2.2. Its impact on SKA development was different in character from ASKAP and is described in Sect. 6.2.2.5.

The Low Frequency Array (LOFAR) began as a named project in 2000 with a paper by Jaap Bregman (ASTRON) (Bregman, 2000), although development had already begun in 1998 (van Haarlem et al., 2013) (see Sect. 3.2.6.1). The Murchison Widefield Array (MWA, see Sect. 4.3.3.1) arose in about 2004 because of a split among the original promoters of LOFAR about its location. The radio astronomy group at MIT/Haystack led the initial design of the MWA (Lonsdale et al., 2000) (see also Sect. 6.5.3).

By 2006, sufficient progress on low-frequency arrays was made that further development specifically for the SKA was not considered important (i.e., in the context of the Reference Design or Memo 100 (Sects. 6.2.1.3 and 6.2.1.4)). It was also thought that design choices, which might have been controversial, were not needed.

6.2.1.6 Aperture Array Development Programs for the SKA: SKADS and AAVP

After several years of planning, the SKA Design Study (SKADS) was funded for 4 years in July 2005 through the European Commission’s Sixth Framework Program (FP6) for research, technological development and demonstration (see Sect. 3.3.3.4) and from national sources. As outlined in Chap. 4, the EC funding had a major unifying effect among the European SKA participants, and through wide international participation, provided significant momentum for the whole SKA project and significant technical training opportunities for young researchers. An outline of the technical progress made in the SKADS program is contained in Sect. 6.5.5.2.

It was clear in November 2009⁴⁵ that many challenges remained before dense AAs could be adopted for the SKA. Hence, development continued beyond SKADS, as dense AAs became the focus of the Aperture Array Verification Program (AAVP) (coordinated by ASTRON and funded partly by a consortium of mainly European institutions plus ICRAR in Australia and partly by the PrepSKA program⁴⁶). Discussions of the plans to establish the AAVP as a formal collaboration independent of, but associated with, PrepSKA were already being held at the SPDO offices in Manchester in 2008.⁴⁷ Work formally began at a meeting in Zaandam in March 2010,⁴⁸ shortly after SKADS funding ceased.

6.2.2 2008–2012: The SPDO Period

PrepSKA funding was a step change in the SKA project. It provided a serious level of funding on the basis that a buildable design would emerge at the end of three years (later extended to four). Although challenging, science goals were clear and impressive enough to attract the attention of the astronomy world and funding agencies. As already described, engineering and technical progress had been developing in associated institutions globally and had already attracted considerable funding in Europe. The enthusiastic flavour of all this is captured in the January 2008 SKA newsletter.⁴⁹

Now came the hard part, putting together a comprehensive design from everything that had gone before. Previous development progress led to expectations by individuals and institutions that what they had put forward would be integrated into the SKA system design. This integrative approach was embodied in the primary technical work package description in the PrepSKA proposal, Work Package 2 (WP2).

While this was an interesting principle, it turned out to be impossible, especially at the expected scale of the project. To develop a buildable design in four years, decisions and technology selections had to be made. Ideally, decisions would have been made at the engineering level because only at that level can feasibility and cost be properly assessed. In reality, engineering decisions are based on the balance of probabilities because not all information is available, and schedule pressures demand answers. For a project the size of the SKA in which none of the participants had direct experience, a conservative approach would have been warranted, but an overly conservative design would not have yielded the scientific advances required. This created constant tension in design discussions. Although common in many projects, this is a salient feature of complex projects (see point {E} in Chap. 6 introduction).

This section contains the story of the SKA Program Development Office (SPDO) from the engineering perspective, and how the architectural design of SKA Phase1 was finally brought together. The story has many threads, the most important of which concerned the technologies that were finally adopted for the design of SKA Phase 1, dishes and low-frequency aperture arrays. Significant innovations in dish designs were tried and discarded but the project did settle on a design that had not been used previously in radio astronomy.⁵⁰ Similarly, ambitious initial plans for aperture array technologies were pared down drastically but did retain new design aspects. The sagas that led to these results are told in Sects. 6.4 and 6.5.

The project managed to stay together despite clashing ambitions and heart-breaking decisions that had to be made to reach a practical, affordable design.⁵¹ Given the global nature of the collaboration, major PrepSKA WP2 meetings were crucial. They were the glue that facilitated communication and mutual understanding among those working at the technical level.⁵²

6.2.2.1 Inherited Directions from the ISPO and Grasping New Challenges

In 2007, the view on engineering and technology expressed in the PrepSKA proposal⁵³ was that “a Central Design Integration Team (CDIT) will be formed, … This team will have as its primary task the goal of integrating all the diverse strands of technology development from around the around the world to produce a detailed and fully costed design for Phase 1 of the SKA, and to develop a deployment plan for the full SKA.” The approach was perfectly illustrated in Fig. 6.2 from the proposal and Memo 100.

Fig. 6.2

A schematic diagram showing the central role to be played by the ISPO-CDIT in taking the technology innovation and prototyping carried out by the design studies like SKADS in Europe and TDP in the USA and the pathfinder telescopes (ATA, EVLA, e-MERLIN, LOFAR, APERTIF, MeerKAT, and MIRA) to an integrated end design for the SKA (SKADS: SKA Design Study, TDP: Technology Development Project (USA), ATA: Allen Telescope Array, EVLA: Expanded Very Large Array, e-MERLIN: enhanced Multi-Element Radio Linked Interferometer Network, LOFAR: LOw Frequency ARray, APERTIF: APERture Tile In Focus (an upgrade of the Westerbork Synthesis Radio Telescope), MeerKAT: enhanced Karoo Array Telescope, and MIRA: Mileura International Radio Array). Credit: Peter Hall

There was already a Reference Design (Sect. 6.2.1.3) and top-level specifications in Memo 100 (Sect. 6.2.1.4). Also, large-scale design studies⁵⁴ were in train or being planned, and numerous pathfinder and precursor telescopes⁵⁵ were committed to providing design information to the project. This was a good base from which to begin.

Specific objectives of WP2 were to produce (see also Chap. 4, hba.skao.int/SKASUP4-4):

• (a) A costed top-level design for the SKA, and a detailed system design for SKA Phase 1,

• (b) Advanced prototype SKA sub-systems specified as part of (a), the sub-systems to be based on technology development in the current regional pathfinders and design studies,

• (c) Base receptor technologies for SKA Phase 1 and critical wide field-of-view design technology extensions,

• (d) An Initial Verification System (IVS) which rolled together the most advanced SKA Phase 1 technology components and demonstrated the functionality, cost effectiveness and manufacturability of the adopted SKA Phase 1 design.

Although this was a very wide scope, PrepSKA managed to accomplish a significant fraction of these objectives. But towards the end of the PrepSKA period, it became clear that another phase of work (SKA Pre-construction) would be needed before readiness for construction could be demonstrated.

6.2.2.2 Engineering Management: Changes in Direction

While the PrepSKA proposal clearly laid out the top-level organisation of work packages, including a breakdown of PrepSKA WP2 into smaller ‘programs’ (see Fig. 5.23), the first job of the SPDO was to analyse them and provide a plan for actually delivering a costed design using the available resources. As explained above, the general approach emphasised integrating all the diverse strands of technology development from around the world into a telescope design.

Over the ensuing months after the new Project Engineer, Peter Dewdney, joined the SPDO in April 2008, analysis revealed practical difficulties in managing the integration process described in the PrepSKA proposal. The root issue was that the SKA participants had long agreed on ambitious scientific goals and had each been working on new technology solutions to enable them at modest cost.⁵⁶ But just integrating them into a coherent system was clearly going to be impossible without making some changes in the approach at the top engineering level.

While the previous ‘integrative approach’ never reached extremes, the greatest fear was that everything would be cobbled together with substantially different approaches to engineering and bespoke interfaces. This is epitomised in Fig. 6.3, used at the time to illustrate the point, for example in October 2010.⁵⁷

Fig. 6.3

An illustration of the complexity of stitching together devices which were developed with disparate interfaces and standards. Credit John T. McFarland

The answer was to adopt a more formal system-engineering structure that would stand the test of time, enforce documented/verified decision-making and reviewed milestones, and provide a structure that could accommodate the scale of the SKA project. This last item was especially important because no one in the radio astronomy community had real experience with projects of this scale and access to this sort of expertise was limited. Scale changes everything!

This was the SKA response to two of the issues raised in the chapter introduction (see points {J} {K} in Chap. 6 introduction): The need to deliver and Project Management and System Engineering. To avoid the tunnel-vision effect noted in {K}, it was important to avoid an overly restrictive form of project management. Several promising innovative, but high-risk, approaches were kept on board, but subject to standardised reviews.

A key aspect of the system-engineering approach was deriving technical requirements from science requirements, which is covered in Sect. 6.2.2.7. While previously this had been done informally in SKA Memo 100 (see Sect. 6.2.1.4) and other places, a more precise approach was needed. Formal experience with system engineering existed in the South African engineering community and had been employed in the design and construction of the South African Large Telescope (SALT), an optical telescope. With South Africa being an active partner in SKA development and vying for the SKA site, there was a ready-made opportunity to exploit their experience. It was therefore not surprising that the SPDO System Engineer appointed in 2008, Kobus Cloete (SKA South Africa), came from South Africa.

The SKA project certainly followed the definition of complex project as defined by Sanders,⁵⁸ so complex that its prospects were considered extremely uncertain. This was all the more reason to adopt a more rigorous approach where possible.

Thinking theoretically along system engineering lines was a long way from convincing the participating institutions to follow the approach or to develop a detailed plan. An analysis of the tasks and how each participating institution was planning to contribute led in November 2008 to the Guiding Principles document.⁵⁹ The analysis breakdown is illustrated in the figure in SKASUP6-1,⁶⁰ in which all 40 tasks were organised into Programs (P1–P10), and each task was managed by a Lead Institution, leading as many as nine contributing institutions for a given task.

Some important observations as quoted from the Guiding Principles document were:

"The work involved is complex for four reasons:

• 1. The project contains research elements and unknowns and is not like any projects that the contributors have carried out before.

• 2. There is a highly fragmented network of contributors, even to the individual task level.

• 3. The locations of the contributors are globally dispersed.

• 4. The resources of the SPDO are insufficient to take up any slack left by the contributors."

"It is obvious by inspection of the task grid⁶¹ that while the tasks are challenging in themselves, the project is additionally complex because of the number and diversity of contributors. In some cases, there are so many institutions involved in some tasks (e.g., Tasks P9T5 and P9T6, which each contain ten contributors) that communication may involve more effort than the work."

The complexity of the network of contributing institutions was brought about because most of the institutes had some expertise over a broad range of skills and wanted the chance to contribute, but also to get as large a slice of the project as possible. Also, the SPDO was new in 2008 and needed time to gain the trust of experts in the contributing institutes. Each thought that they should be involved in almost everything. Clearly this is also a hallmark of a complex project (see point {E} in Chap. 6 introduction).

Some of the complexity was foreseen in the original PrepSKA proposal, which suggested the appointment of Liaison Engineers in the institutes (see SKASUP6-2). They would have been required in any sensible organisation with many contributing organisations but were also helpful in mitigating the obvious management complexity described above. In the Guiding Principles document, they were described as “senior engineering managers, part of normal management structure—not necessarily experts in the technologies associated with each task, but sufficiently knowledgeable to understand and support multiple tasks if required.”

As described in Sects. 3.3.1.3 and 4.4.2.1, the previous engineering working group (EWG/IEMG) continued during the PrepSKA period, alongside other working groups (e.g., Science WG, Site Characterisation WG, Simulation WG, Operations WG, Outreach WG), each with its own chair.⁶²

The Guiding Principles document and a draft Project Management Plan were in place for the PrepSKA WP2 Kick-Off Meeting held in Manchester in November 2008.^63,64,65 This was an optimistic time for the project. Individual participants were pleased that their home institutions were supporting the project and that most would be able to continue their own participation. An initial plan had been circulated from the SPDO, so that most participants had a good idea of their roles. SPDO had hired a project engineer, a nearly complete complement of domain specialists, a project scientist, a system engineer, and a site engineer. In addition, the Project Scientist had already introduced the first cut at the development of science requirements.⁶⁶ Nevertheless, participants had doubts about the change to a system engineering approach, doubts which persisted for a long time afterwards and never really went away. Final development of revised work plan⁶⁷ took another year to be finalised, an indication of the length of time needed to change direction.

6.2.2.3 International Engineering Advisory Committee (IEAC)

The need for an engineering advisory committee had already been realised and its terms of reference for the IEAC were first drafted for the ISSC meeting in Guiyang (2005) and subsequently revised.⁶⁸ Its purpose was to provide the ISSC (subsequently the SSEC) with external expert advice on SKA technical progress by annually reviewing past engineering progress, and near and longer-term plans. Members⁶⁹ were selected to cover the technical breadth of the SKA. The IEAC was asked to review documentation provided by the SPDO, reports from design reviews, project management reports, presentations on up-coming technical issues, and reports from precursor and pathfinder projects.⁷⁰

The report from the first meeting in April 2009 was prescient in many ways. Almost all the findings⁷¹ continued to be issues throughout the PrepSKA period and beyond: for instance, very challenging timescales, efficiency in coordinating a globally distributed engineering project, and the need for consensus from the institutions and regional development groups on rules for PrepSKA decision-making. On the more technical side, obtaining acceptance of the proposed dish verification program (DVP), coupling between the design and the choice of site, a method for ranking the pathfinders for performance/cost ratios (emphasising sensitivity and controlling systematic errors), and isolating critical technology path(s), and figuring out how to proceed to SKA Phase 2.

The second meeting⁷² (June 2010) was held after the System CoDR (see Sect. 6.2.2.9). Although they endorsed the SKA Phase 1 concept definition following the CoDR, they noted that a System delta-CoDR would be required and strongly recommended that “any further changes to (the) concept definition be made on the advice of the SPDO, after due analysis of performance—cost considerations and other engineering issues.”

The third meeting⁷³ (June 2011) noted additional progress, especially in executing several sub-system CoDRs, areas of new governance structure, selection of a new headquarters location, site selection progress and the review of the Project Execution Plan (PEP))⁷⁴ (see Sect. 6.2.2.14).

This committee, now the SKA Engineering Advisory Committee (SEAC), continues with a similar mandate at the time of writing.

6.2.2.4 Recruitment and Staffing the SPDO

There was great concern among some participating institutions that recruitment at the SPDO would rob them of their best talent, hobbling their other activities and rendering them less competitive in promoting their favourite technology. In other cases, they were simply concerned about growing requests for funds over which they might have little control. The directors of these institutions and some of the funding agencies represented in the FAWG and later the ASG⁷⁵ made it clear that the SPDO had to be a small group who could design a system from what they had to offer.

Although this could have been a disaster, it did not turn out that way. There were some advantages to recruiting people from outside the radio-astronomy field, who had experience in large, international projects (e.g., potentially from large companies) to balance those from the relatively small numbers of people in the field globally. In other words, it was difficult (or unrealistic) to find persons with long experience in radio astronomy technology who also had large project experience. Some existed but were not available because they were well established elsewhere or for other reasons.

The terms of employment were a big impediment. Applicants had to move to Manchester⁷⁶ for a job with only a four-year term, and if the project could not continue, and if they were not EU citizens, they would have to leave the UK.

In the end, most came from the UK and the others from the participating countries. The result was a staff complement with a mixture of backgrounds, and several without strong connections to radio astronomy. It was gratifying to be able to find people who believed that scientific progress also meant taking risks with their careers. SKASUP6-2⁷⁷ contains a list of SPDO staff from 2008 to 2012 (see also Fig. 4.6).

Another problem was simply one of numbers. The complex nature of the SKA project (see Sect. 6.2.2.2 and Sect. 11.3.2) required a larger staff complement to manage and coordinate activities than could be funded at the central office. Secondment from participating organisations was an alternative but this did not materialise to any substantial extent. Although the Liaison Engineer solution did help, it was not sufficient to manage the project effectively. Liaison engineers, themselves, had ‘day jobs’ in their home institutes and were not always able to spend the required time on the SKA.

Nevertheless, with a few bumps in the road and some muddling through, most of the SPDO staff continued to the end of PrepSKA and some went on in 2012 to the next stage in the SKA saga, the Pre-construction era. A sense of optimism (see point {A} in Chap. 6 introduction) and a belief in the project’s vision, to build the world’s largest radio telescope, carried them through (see Sects. 11.3.5 and 11.3.6).

6.2.2.5 Other Challenges

The impact of SKA Pathfinders and Precursors⁷⁸: These projects played a complex role in the SKA’s engineering development with both positive and negative effects in the 2006–2012 period. On the positive side, the SKA provided motivation for the participating countries and organisations to develop their own expertise and projects in what was seen as an important field of astronomy. The development of expertise in particular had a lasting effect on the SKA’s prospects at the end of PrepSKA, as it entered the pre-construction and construction phases.

On the other hand, in the participating countries, the most likely route to successfully raising national funds was to propose institute or national SKA-related projects, in addition to annual contributions to the SPDO operating costs.⁷⁹ National and institutional funds were then spent locally. With the consequent requirement to satisfy local deliverables, there was a tendency for national priorities to take precedence over the global project. From a central project office perspective, a less than positive side-effect was to utilise people and resources that otherwise could have been directed in a more-focussed way towards the global project itself.

Throughout PrepSKA, the (perceived winner-takes-all) competition for the SKA site created a hyper-competitive atmosphere in the project that had its effects on the large precursor projects, ASKAP and MeerKAT. Statements that carried an implication that a particular technology being developed in one or the other precursor might already be discounted or left out were very sensitive.

The large precursor telescopes (ASKAP and MeerKAT) impacted the SKA project in a similar way as noted above. But their motivations were completely different⁸⁰:

• Australia had a long, storied history in radio astronomy going back to the beginning, having already built a suite of scientifically productive telescopes. Therefore, the motivation for ASKAP was to build on that past and to progress the field in a fundamental way. It was initiated as a vehicle for technology development and to showcase Australia’s position as a world-leading supplier of innovative technology for radio astronomy and the SKA. But it later became a showpiece to highlight the Australian site and a fall-back if its site were not selected.

• South Africa heretofore had only a small role in radio astronomy, with one small telescope, used primarily for VLBI. But South Africa had selected astronomy as one of four science fields to emphasise⁸¹ (see Sect. 3.3.3.7) and needed a project to show the world that South Africa could build a major radio telescope as more established countries had done in the past. Although not really a vehicle for technology development, they did this in spades, at the same time highlighting the deep reservoir of engineering talent in the country. As the site competition heated up, MeerKAT also provided a fall-back position for a role in radio astronomy, whatever the decision.

Cultural Differences: In general, different styles of work in different organisations, one of the aspects of a complex project, is inevitable but does make it difficult to design an integrated project (see point {E} in Chap. 6 introduction). Contrasts between the engineering and science cultures were evident between Australia and South Africa. Just as great were contrasts in approach between Europe and the USA. The European approach of consultation and compromise, to counter a legacy of centuries of war, was partly responsible for keeping the SKA project afloat through formative periods (see Chap. 4). In the USA, the more competitive and combative approach was partly responsible for its withdrawal from the project in 2011 (see Sect. 4.5.3). Both have their strengths and weaknesses, but in the case of the SKA, the European approach was far more productive.

Communication and documentation: Communication in a such large project, whose participants were spread over several time zones, was always going to be a challenge. The annual large engineering meetings (see Sect. 6.2.2) were critically important in maintaining contact among the community of engineers across the globe, but a year is a long time between updates in a four-year project.

One remedy for communication challenges was thought to be good written documentation that could be reviewed and passed back and forth among participants, ultimately becoming the only source of corporate memory. The SPDO developed detailed standards for documentation.^82,83 However, many of the institute laboratories were not used to writing project documentation, often resorting to Power Point presentations at meetings with little follow-up. Academic institutions are motivated by career progression and grant applications to produce peer-reviewed papers in journals. However, these usually take months to become available and are much less detailed than project documents. In the development stage of a project, it is important in documentation to relate what did not work, not just what worked.

In retrospect, the project’s expectations were probably too high. Detailed documentation takes a great deal of work to produce. Despite these shortcomings, when it came to design reviews, the documentation produced for them provided a sufficient legacy to allow the project to track progress for several years and eventually to make sensible decisions.

The Effect of US Withdrawal: Scientists from a broad cross-section of US astronomy were leading participants in the SKA project until the 2010/11 era and played a key role in developing the technical definition of the SKA in its early stages and during the first part of the PrepSKA era.

When the report from the ASTRO2010 Decadal Survey of astronomy (Blandford et al., 2011) contained only mild support for US involvement in the SKA, the National Science Foundation (NSF) declined to participate further (see Sect. 4.5.3). Very high cost-estimates from the Aerospace Corporation (see Sect. 4.5.3.5), which have never been revealed in detail, also played a role. The consequence was that the deep reservoir of engineering and scientific talent in the USA was ultimately lost to the project.

As US astronomers were primarily interested in astronomy at the high end of the frequency range of the SKA, they concentrated mainly on defining SKA-Mid. The NSF-funded Technology Development Program (TDP) was the primary vehicle for supporting dish development. Its impact on dish development is discussed in detail in Sect. 6.4.5.1. The US expectation for the SKA was to build a third telescope in the USA for frequencies well above 10 GHz, which was then referred to as SKA Phase 3.

Although opening new discovery space (see point {B} in Chap. 6 introduction) (e.g., in the time domain) was not a major part of the SKA science case, it did play a role, denoted by Exploration of the Unknown (see Sect. 6.2.2.8). However, the US funding system is not receptive to this (see Sect. 4.5.3.4) and building flexibility into a design for the sake of enhancing discovery potential was not seen as a valid argument, as discussed in Sect. 4.5.3.4. Nevertheless, it was mainly US astronomers who promoted discovery space in the time domain.^84,85

6.2.2.6 Design Reviews

The PrepSKA design review process was not particularly new or innovative, but it was relatively new to radio astronomy projects. Design reviews are a formal process that brings together all the aspects of a project as it moves through various stages. For PrepSKA, this was spelled out in the System Engineering Management Plan.⁸⁶ Figure 6.4 is a simplified view of how technical options were to be narrowed down through a series of reviews. An initial set of system requirements were developed,⁸⁷ and plans were made for a System Requirements Review (SRR),⁸⁸ intended to ensure that requirements were well understood and suited to the project. Project review practice varies, but during PrepSKA most of the reviews were conducted at the Concept Design Review (CoDR) level. The PrepSKA review process required between 10 and 20 documents for each review.

Fig. 6.4

A schematic view of the reduction of potential technical options available, through a series of reviews

Supplementary material⁸⁹ provides details on how the reviews were carried out.

6.2.2.7 Developing Technical Requirements from Science Requirements

The work of the URSI Large Telescope Working Group (LTWG, see Sect. 3.2) in the 1990s was devoted to developing the initial science requirements for the SKA and discussing the technical specifications to deliver the science. In 2001, based on this work and other national efforts, Ron Ekers summarised the then current technical specifications for the SKA.⁹⁰ This was updated by Dayton Jones (Jet Propulsion Laboratory) in 2004⁹¹ (see also Sect. 5.7).

One of the first orders of business for PrepSKA was to extend the work of Memo 100 (see Sect. 6.2.1.4) and produce more formal connections between the science goals of the SKA, and the technical requirements which embody the scope of the telescope design in top-level terms. This was easier said than done because for a general-purpose telescope like the SKA, the science is open-ended and diverse. This contrasts with some astronomy projects, typically space missions, which are designed to answer a small number of important, well formulated science questions and no more.

A school of thought circulating at the time was: “Why not just improve the capabilities of each generation of telescope along the ‘fundamental performance axes’: resolution, sensitivity, frequency coverage, time-domain resolution, field-of-view, polarisation and spectral coverage/resolution?” But which axes to choose was the question? A partial answer lay in the concept of the SKA in the first place, “Improve sensitivity by two orders of magnitude!”. This was partly because in radio astronomy, very high-resolution imaging through Very Long Baseline Interferometry (VLBI) had already been achieved (see Sect. 2.3), demonstrating that it was possible to build telescopes with effective diameters up to near-Earth orbit, and routinely the diameter of the Earth. With the collecting area available there was only enough sensitivity to detect extremely bright non-thermal objects, principally quasars.

Still a debating point, another fundamental performance axis was frequency range. The early proponents of the SKA were mostly interested in decimetre wavelengths and longer. But the SKA also attracted a large contingent of potential observers that wanted centimetre wavelength coverage.

Naturally, cost and technical feasibility (see the green box in Fig. 6.5) constrained the design process. To obtain tangible information in these two areas, the key line in Fig. 6.5 is Case Studies. Scenarios were selected to capture the upper performance envelope that would progress existing science in directions commensurate with SKA science goals. These provided the guidance needed to incorporate that capability into the design requirements (see SKASUP6-4⁹² for further discussion). Cost models of the entire SKA system are also discussed in Sect. 6.4.6.

Fig. 6.5

A diagram used in a presentation by Dewdney at the SKA Forum in Perth in 2008 (hba.skao.int/SKAHB-154 The SKA—Opportunities & Challenges for Industry, Dewdney, P. E., presentation at the SKA Forum in Perth, April 9 2008) to represent the approach to deriving a design from science requirements

At a high level, SKA science had long been targeting Key Science Projects) (KSPs),⁹³ broad categories of science for which radio astronomy could deliver unique or complementary results to all of astronomy. Case studies were particularly important in these areas. This began with a draft set of such studies in 2008⁹⁴ and a presentation by Joe Lazio, the new SKA Project Scientist, at the PrepSKA kick-off meeting.⁹⁵

This process cannot operate without feedback from the prospective observer community. The SPDO and the Science Working Group developed a Reference Science Mission (RSM) through 2009,⁹⁶ and in 2010 this document became the Design Reference Mission (DRM).⁹⁷

Although the DRM provided guidance to the design, a simple flow-down of requirements to technology selections was not possible in 2010. The SKA project was not ready to permit a simple approach to selecting technologies for the design (i.e., establishing a design baseline). However, this was established in March 2011, a month after the System Concept Design Review (CoDR) (see Sects. 6.2.2.9 and 4.5.2).

In parallel with preparations for the System CoDR, a method to guide the selection of technologies was developed using an ‘evaluation hierarchy’ to evaluate potential SKA system implementations.^98,99 (See the second figure and the discussion in SKASUP6-4¹⁰⁰).

6.2.2.8 Exploration of the Unknown

A historically important goal of the SKA was “The Exploration of the Unknown”. Most of the major discoveries in astronomy have been unexpected or accidental, relying on the perspicacity of observers and some luck. There is no design method that can uniquely capture this goal. Informed judgement plays a key role on whether to spend resources on a design aspect that could convey additional design flexibility or agility to widen discovery space (see point {B} in Chap. 6 introduction). An important early analysis of discovery space was described by Jim Cordes,¹⁰¹ in which various aspects of discovery potential are discussed. In particular, he illustrated the size of the time-luminosity phase space covered by the relatively small number of known transient phenomena, about 20 orders of magnitude on each axis.

An important aspect of the SKA’s approach to discovery space was to ensure continuous frequency coverage within the overall boundaries set by the major goals of the telescope. In the case of the Expanded Very Large Array (EVLA) for example, continuous coverage had a dramatic influence on the number and quality of new phenomena (discoveries) made after the VLA was equipped with receivers that covered its entire accessible frequency range.

A design aspect related to discovery space is ensuring continuous, smooth coverage of telescope sensitivity over its range of accessible scale sizes. During the development of the SKA array configurations, there was much discussion of this point (see Sect. 6.2.2.10). A related concept is spatial dynamic range, which is a measure of the range of scale-sizes on the sky that can be recovered from the observations. As an example, for some types of radio sources, a so-called wide-shallow survey of a large area of sky can yield more discoveries than a survey with higher sensitivity on a smaller area of sky, in the same amount of observing time.¹⁰² All these aspects are ultimately linked to survey speed (see Sects. 6.4.1 and 6.5.1).

Historically, the design or operation of many telescopes have effectively discarded discovery space by, for example, using long integration times, imaging over a narrower field than the antenna beam or simplifying polarisation observations. The design of the SKA was intended to capture all the available information from the telescope so as to maximise discovery space. Further examples of this are given in SKASUP6-5.¹⁰³

6.2.2.9 The System Concept Design Review (February 2010)

A major transition during PrepSKA occurred because of the System Concept Design Review (CoDR). Up to that point, the only accepted view of the telescope design included ‘everything’. Figures 6.6 (left) and 6.7 (right) illustrate the scope of the issue.

Fig. 6.6

A compact system view of the SKA telescope. Left: The technologies associated with the SKA before the System CoDR. Right: The technologies associated with the SKA after the System CoDR and the definition of SKA Phase 1. Credit: The authors using material from Swinburne University of Technology, Melbourne, Australia

Fig. 6.7

The proposed layout on the ground of the central part of the SKA site for the original SKA (right) and SKA Phase 1 (left). Note the presence of antenna clusters rather than single antennas

In late 2009, frustration had been building at the SPDO (see Sect. 4.5.2). It was proving very difficult to select technologies, regardless of the rigour of the design process (see Sect. 6.2.2.7). Each of the national participants strongly believed that their approach to technology was the best. It was also evident that managing the project and completing its goals would likely be impossible without acceptance by the participants that choices had to be made to achieve the required focus.

It became obvious that the only way to proceed was to hold a review with senior independent experts on the Review Panel who had experience in similarly large (or larger projects), where technology had been selected.¹⁰⁴ Although this might have put the project at risk of cancellation if the review went badly, it was felt that in the medium term the project was at risk anyway.

A three-day CoDR meeting was organised in mid-February 2010 around a review plan.¹⁰⁵ The review was based on a comprehensive set of documents, provided by the SPDO and national participants, which the latter had an opportunity to review before they went to the Review Panel.

The key documents were the Design Reference Mission,¹⁰⁶ the High-Level System Description,¹⁰⁷ the Science Operations Plan,¹⁰⁸ Strategy to Proceed to the Next Phase,¹⁰⁹ the Risk Register,¹¹⁰ and the Risk Management Plan.¹¹¹ There was also a presentation on cost, but because of the uneven development states of the technologies involved, a definitive report on cost was difficult to produce.

As expected, the Review Panel found that the project was not ready to proceed. Among the key findings^112,113 were:

• SKA in its present setup tries to push technology limits on pretty much all fronts. Some parameters are pushed orders of magnitude beyond state-of-the-art. Even things that traditionally have been minor problems are now an issue (e.g., power, computing, signal transport and processing, etc.). Given current time and cost constraints the Review Panel felt that the combination of scope, timeline, and cost was in general overambitious and in several areas unrealistic.

• Given current timeframe and assumed funding constraints, the science covers too large a parameter space and includes requirements which imply differing optimal design decisions, … .

• The Review Panel did not see stable requirements which would allow a stable design for SKA.

• SKA is ready to move into the definition phase (Fig. 6.4). This transition is essential to support the proposed timeline for a construction start (with a redefined scope), to arrive at an SKA concept, and to ensure that additional resources are focused on activities that truly support the SKA schedule.

As explained in Sect. 4.5.2, the impact was immediate. In response,¹¹⁴ the SKA Science and Engineering Committee (SSEC) convened a sub-committee to define the science goals and a concept technical baseline for SKA Phase 1 with the aim of stabilising the design requirements as soon as possible. In other words, detailed design work would proceed for SKA Phase 1 only. The sub-committee produced a brief, influential report, SKA Memo 125,¹¹⁵ that set a new direction for the SKA. It refined the science goals, the technical baseline, and cost and schedule targets. It also established an Advanced Instrumentation Program (AIP), which provided a mechanism to continue to develop less mature technologies (Phased Array Feeds, Wide-band Single Pixel Feeds and Aperture Arrays, see later sections in this chapter) in the expectation that they could be used in SKA Phase 2. The new approach was embodied in the Project Execution Plan (PEP))¹¹⁶ generated in late 2010 (see Sect. 4.5.3.7). It retained the momentum of the project and the need to deliver (see point {J} in Chap. 6 introduction). The resulting system is depicted in Figs. 6.6 (right) and 6.7 (left).

The technical baseline consisted of low-frequency sparse aperture arrays and a dish-array of 15-m antennas equipped with single-pixel feeds. However, it did not entirely stick because, as a part of the site decision process in 2012 (see Sect. 8.6.3), an additional array of dishes with Phased Array Feeds (an AIP element) became included in the technical baseline until it was removed in 2015.

As a result of Memo 125, the SKA adopted mainly proven technology for the SKA Phase1 technical baseline. This was the moment which inspired the title of this chapter: “Innovation Meets Reality”.

6.2.2.10 Intertwined: Array Configuration, Infrastructure and Topology

Even after a suitable site has been identified, the design of a large telescope array like the SKA requires consideration of many interacting, practical factors. Antennas must be placed on solid ground where service access is available, where the sky is not blocked by local topology, and in the case of antennas far away from the core, where the climate is acceptably benign. Cost plays a significant role. In contrast, the ideal array configuration from the perspective of telescope performance is not likely to be compatible with these practical constraints, and a compromise must be reached. Figure 6.8 shows the many aspects considered in the design of the SKA array configuration.¹¹⁷

Fig. 6.8

A context diagram showing all the influences and aspects of the design of the SKA array configuration

An array configuration based on a spiral pattern on the ground had been accepted as the best available basic configuration for the SKA, although previous major telescope arrays had used a Y-shaped configuration. SKASUP6-6¹¹⁸ explains how the spiral configuration influences telescope performance and the factors used in assessing performance. One of the dominant factors in deciding on the configuration was the cost, which was in large measure dependent on the number of spiral arms and number of “stations” (assemblies of multiple antenna elements) along the arms.

A Configuration Task Force (CTF)¹¹⁹ was set up by the SPDO in April 2008¹²⁰ to optimise the array configuration, including “matching the ‘ideal’ configuration to the geographical realities of the two short-listed sites”¹²¹ (see SKASUP6-6 and Sect. 8.4.4).

Once a spiral approach was adopted, there were not many avenues of optimisation of the spiral, itself: the number of arms, the wrap of arms, the spacing of antennas on the arms, and the fraction of antennas in the core. For example, once cost minimisation is considered, the number of arms is affected. Figure 6.7 (right) shows a five-arm spiral, which is better than a three-arm spiral, but costs more. But was a three-arm spiral good enough? This could only be judged against the science goals. A prominent cost-driven issue was whether antennas could be clustered on spiral arms, rather than spreading them out along the arms.¹²² This would reduce servicing cost but produce redundant samples in the u-v plane, instead of providing more distinct samples.

Collaborating with the Science Working Group, the Simulation Working Group, and other members of the SPDO, the CTF adopted figures-of-merit (FoM), based on earlier work,¹²³ for evaluating the imaging capabilities of proposed configurations after they were subjected to some of the practical site constraints described above.¹²⁴ The SPDO also produced a model for the site selection process¹²⁵ in 2011–12, which contained considerable technical information on site requirements (see Sect. 8.3.7).

Once the dual site decision was made in 2012, the job of optimising the SKA array configurations became much easier, but this was not finalised until after 2012. A UK branch of the company, Parsons-Brinckerhoff,^126,127 with experience in large infrastructure projects was commissioned by the SPDO to assess the feasibility and cost of the proposed infrastructures on the two sites, based on the model configuration in the Request for Information document (see Sect. 8.3.7).

6.2.2.11 Site Power Provision

The cost of supplying electricity to a remote SKA site had been recognised from the beginning as an important factor in the SKA capital and operating costs. In February 2009 the SPDO set up a Power Investigation Task Force (PITF), co-chaired by Peter Hall (Australia) and Bernie Fanaroff (South Africa)¹²⁸ to assemble experts and understand the impact on power requirements of various possible telescope technologies.^129,130

There was much discussion of ‘green energy’ for the SKA at the time, and various emerging industries were interested in providing both solar power arrays and battery storage. However, the SPDO was concerned that this could be a major distraction from the focus on the science and the design of the telescope, and these initiatives were not taken up at the time.

Radio Frequency Interference (RFI) from power systems had been a longstanding concern. In South Africa, where inexpensive power was available from the grid, the MeerKAT group developed methods for building low-emissions power lines.¹³¹

The potential cost of power distribution on the extended sites engendered discussion of clustering or clumping of dishes,¹³² as well as generating power at remote sites instead of distributing it via power lines. Clustering was eventually taken up for the low frequency telescope in Australia, but not for the mid-frequency telescope in South Africa.

In September 2011, the SPDO took formal responsibility for interfacing with the power industry.¹³³ This work was then led by Phil Crosby, who was already the SPDO Industry Participation Manager.

The on-going cost of power in the operating phase might have been a major limiting factor on just how much sensitivity and angular resolution could have been afforded for the SKA, because providing power to distant antennas is very expensive if it is not available locally. But, at the time of writing, the SKA ‘power problem’ is gradually resolving itself as SKA Phase1 construction proceeds. The emergence of relatively low-cost solar power and rapid reductions in the cost of battery storage, have come to the rescue. Even in South Africa, where grid power has been unreliable for many years, locally produced solar power is likely to be competitive with grid power.

6.2.2.12 Radio Frequency Interference and Mitigation

Radio frequency interference (RFI) from human-caused radio emissions has a strong negative impact on ground-based radio astronomy. Man-made radio emissions, the sources of RFI, have always been a driving factor in locating radio observatories on sites that are as remote as possible from population centres. Figure 6.9 is a dramatic illustration of the efficacy of locating radio telescopes in remote regions. Remoteness is clearly the first line of defence for a ground-based radio telescope.

Fig. 6.9

RFI emission spectrum measurements at three locations in Australia. Left: frequencies from 70 MHz to about 1800 MHz. Right: frequencies from 1000 to 3000 MHz. Note that the vertical scale is different in the left and right panels. Top: Sydney (population, four million). Middle: Narrabri near the Australia Telescope (pop. 6000). Bottom: Mileura, Western Australia at a remote station (pop. 4). (Measurements were made with an HF Coaxial Dipole, omnidirectional, vertical polarisation). Credit: CSIRO Radio Astronomy Image Archive CRAIA-SKA007

While there are several other technical and scientific aspects to choosing a site for a new telescope (see discussion in Sects. 7.3 and 8.3.7), remoteness is a vital factor, which comes at a substantial cost for the construction of the observatory: infrastructure, operations staff, and electricity. The cost of establishing the SKA in remote locations represents a large fraction of the total cost. Every radio telescope in history has had to deal with this issue.

However, the benefits of a remote location in terms of its RFI environment is not guaranteed. Over the years, as previously remote sites have been encroached upon by population centres, operators of radio telescope sites have gradually been able to obtain legal recognition of ‘radio quiet zones (RQZ)’,¹³⁴ which provide a modest level of protection in the long term (see point {I} in Chap. 6 introduction and the discussion of mitigation below).

While astronomy in general benefits greatly from the march of technology (see point {H} in Chap. 6 introduction), radio astronomy is so greatly affected by the growth of radio-spectrum usage that observations will be increasingly difficult. This is the most prominent example of unhelpful technology diffusion for radio astronomy (see point {I} in Chap. 6 introduction).

In particular, there has been a recent proliferation of hundreds to thousands of low-Earth orbit satellites that provide internet and other services to almost arbitrary locations on the ground. Although space-based emissions have been present for decades in bands allocated for radio astronomy (e.g., Argyle et al., 1977), these space-based emissions, even outside the radio astronomy bands, are the most harmful form of RFI. At 11 to 12 GHz, radio astronomy observations are already impaired.

A report commissioned by the SPDO in late-2011,¹³⁵ as part of the site selection process (see Sect. 8.3.6), provided a look at the long-term evolution of the RFI environment generally in the two site countries, considering the increased uses of the spectrum and the intensification with population growth. The main conclusion was that the use of the spectrum in the SKA bands will gradually increase worldwide, affecting both Australia and South Africa.

Aperture synthesis telescopes, consisting of arrays of antennas, are fundamentally more resistant to ground-based sources of RFI because they employ correlation. RFI emanating from sources near the horizon tend to de-correlate, reducing their effect, especially between antennas that are far apart. For example, this effect was invoked in discussions of RFI masks on potential SKA sites.¹³⁶

In the rapidly changing RFI environment, mitigation strategies were recognised as being of critical importance in the SKA design and in the international spectrum management sphere. That remains the case at the time of writing. Hopefully the technical measures taken in the design, most of which are common to many modern radio telescopes, but critically important for the SKA, will ensure a long future for the SKA. We discuss some of the RFI mitigation strategies in the following paragraphs.

RFI Mitigation

RFI mitigation strategies are employed primarily on two fronts, spectrum management and technical design (e.g., see early recognition of this in (Ekers & Bell, 1999)). Fundamental to this approach is establishing a quantitative reference set of measurements of RFI levels at prospective telescope sites. These can be used to plan technical mitigation measures and to make the national and global spectrum-management cases to protect radio telescopes from other users of the radio spectrum.

Reference measurements were carried out for the SKA by Rob Millenaar, the SKA Site Engineer from 2008 and colleagues; these measurements formed part of PrepSKA WP 3 on Site Characterisation. Knowledge of the RFI environment at each of the SKA prospective sites was one of the most important selection factors for both the shortlisting and site decision stages in 2004–5 (Sect. 7.3.2) and 2008–2011 (Sect. 8.4.2 and SKASUP8-2), respectively. RFI played a significant role in the final site decision as discussed in detail in Sect. 8.6.3.

• 1. Spectrum Management: The ISSC established a task force on spectrum regulation in 2004, which covered issues related to site selection,¹³⁷ but also of a more general nature, such as a fully fleshed-out definition of a radio quiet zone (RQZ) . The need to establish an RQZ to protect the observatory site had already been discussed early in the SKA project.¹³⁸ RQZs must be established through national and local government regulations, and some have been established around existing radio observatories.¹³⁹ The SKA RQZ was expected to have international impact, especially since the SKA was expected to transit national boundaries. Specialised conferences were held throughout this period, which included discussion of the SKA’s relationship with the International Telecommunications Union (ITU).¹⁴⁰ Although the SKA could not be represented officially on international spectrum management bodies during the PrepSKA years, Millenaar maintained a presence, for example, at meetings of the Committee on Radio Astronomy Frequencies (CRAF).¹⁴¹

• 2. Technical Measures: Coping with environmental RFI is one of the largest cost drivers for radio telescopes and has substantial impact on the design of the telescope (see also Sect. 5.8.2). There are multiple dimensions to the RFI ‘space’: radio frequency, signal strength, signal bandwidths, signal direction, signal polarisation, time-duration, repetition rate, and spectrum usage-development, among others. As the sophistication of telescopes increases, for example, to accommodate the astrophysical time domain, the full suite of factors becomes important.

  Among the many technical measures for mitigating RFI, a few important ones are discussed here. These directly impact the cost and capabilities of the SKA.

• Dynamic range of RFI and natural signals. Natural signals from the Universe are typically a million to 100 million times weaker than those in most communication systems. The apparatus in radio telescopes that carries the signals (the signal chain) must be capable of carrying both the strong and weak signals without distortion. In terms of bits, only 2 or 3 bits per sample are needed to retrieve the underlying radio astronomy information in the absence of RFI. The presence of RFI signals requires many more bits per sample to accurately retrieve the radio astronomy signal. The cost of separating the weak signals from strong RFI signals increases directly as the number of bits per sample needed to do this effectively, and as RFI signal levels increase more bits will be required. However, there is probably a bit-limit above which this method is no longer effective, especially if large segments of the radio spectrum contain RFI. Modern telescopes require approximately 10 bits to operate properly in the current RFI environment. Although this approach is effective if a significant fraction of the radio spectrum is free of RFI, the spectrum is expected to gradually ‘fill up’ with human-caused signals.

• Avoiding self-generated interference. Radio telescopes contain the same sort of electronics as most industrial equipment. All devices near the site require very high levels of metal shielding to avoid contaminating the radio telescope site. The SKA has also taken the approach of transmitting signals over optical fibre, which does not generate RFI.

• Development of mitigation algorithms. This large collection of techniques is a combination of exploiting the signatures of RFI signals to remove them and exploiting the nature of interferometers to reject signals arising from directions far away from where the science signals originate (summarised in the early SKA era by Albert-Jan Boonstra¹⁴²(ASTRON)). For example, most RFI arises from the horizon around the telescope sites. Both antennas and processing algorithms can in principle provide suppression in those directions¹⁴³ (e.g., see Sect. 6.5.1 for experiments using Aperture Arrays). Algorithms to take into account satellite orbits and emission spectra so that observations can be ‘flagged’ using this a-priori information will need to be developed further.

Although there are data processing algorithms (Offringa et al., 2010; ITU, 2013) that partly expunge RFI from radio telescope data, they are not as effective at removing subtle effects from RFI as human intervention. This will be impossible for SKA because of the huge data volume. In the future it is likely that machine learning and artificial intelligence (AI) will play a role. ‘Training’ these algorithms to recognise RFI in radio astronomy data is already a research topic at the time of writing.

In conclusion, RFI is complex to quantify and changes rapidly with time. This means the attempt to quantify its impact on the long term and mitigate its effects, while critically important, is fraught with difficulty.

6.2.2.13 Operations Planning

Chaired by Ken Kellermann of NRAO until 2010, the Operations Working Group (OWG) was formed in 2004. Its first report¹⁴⁴ was a sensible combination of NRAO’s views on operations and the experience of other large science facilities already operating. Its primary recommendations are contained in Box 6.1. None of these suggestions were particularly new or unknown to the SKA proponents, but they were time-tested goals for most such projects. Some of them were adopted in more detail in subsequent documents. An exception is the “open skies” policy, which was never adopted by the SKA, despite its long history in astronomy¹⁴⁵ (see Chap. 1 and Sect. 4.2, Box 4.1).

Box 6.1 Recommendations of the Operations Working Group in 2004

• Expending capital during the building phase is well worthwhile if it can save operations funding in the future.

• Adopt a ‘multi-tier’ approach to system support and data reduction, similar to that of the Large Hadron Collider.

• Be wary of cost increases resulting from a system of juste retour that is implemented badly (i.e., “avoids potentially inflated contracts from partners who expect their entitlement”).

• As much as possible, a system should be adopted whereby member countries second their staff to serve under the SKA Director.

• Options for power provision, including solar, wind, and geothermal sources, should also be explored in relation to operational issues.

• Because software development is so expensive it will be important that the resources be made available to make software easily portable to new computing hardware.

• An “open skies” policy with peer reviewed assessment of the scientific quality of proposals will give the best science returns for the SKA.

In 2010 the OWG presented a broad-brush, but comprehensive, plan¹⁴⁶ for SKA operations for the System Concept Design Review (System CoDR). This was an updated follow-up to the principles outlined in Box 6.1. Although it was based on a single-site model with a large, in-country headquarters, many of the concepts have survived to the present. However, as in the previous OWG document, there is a strong emphasis on anticipating and controlling operations costs. The emphasis on governance style, in which the Observatory Director would be vested with full responsibility in day-to-day operations, is less practical for the current multi-site model.

As part of the site selection process, an SPDO analysis of the staff required for operations¹⁴⁷ was included in the 2011 Request for Information from Candidate Host Countries to allow estimates of infrastructure requirements for the full SKA (SKA Phase 2) on a single site by the candidate sites. On-site staff were almost entirely maintenance staff, while observatory operations staff were off-site. Staff numbers were roughly based on failure statistics of components and the numbers of components in the system. The numbers startled many people and led to criticism that there was no way that these staffing levels would be needed, because modern remote operations would be brought to bear on the problem. However, it was more likely an indication of the ambitious scale of the full SKA system. In the future, this will be tested on a small scale with SKA Phase 1 on two sites.

A more forward-looking operations concept document was written for the delta System CoDR, which followed the System CoDR, in 2011.¹⁴⁸ This became the basis for subsequent work and presented a more detailed description of Regional Science Centres.

As a postscript to the long-running development of an operations plan, under the chairmanship of Douglas Bock, the OWG produced a comprehensive top-level operations plan in 2013 that was written to be a parallel to the SKA Baseline Design¹⁴⁹ document. However, the SKA Organisation Board of Directors was not ready for such detail, and asked that it be reduced to a set of operational principles. After many iterations, the principles document¹⁵⁰ was accepted and a revised edition of the more complete document¹⁵¹ was eventually released.

6.2.2.14 Outcomes

As a result of the reports from the System CoDR in March 2010 (see Sect. 6.2.2.9), things moved rapidly over the next year:

• 1. In August 2010, the SSEC revised the overall SKA plan to build the SKA in two phases, outlined major science goals and a technical baseline for SKA Phase 1 in SKA Memo 125¹⁵² (see Sects. 4.5.2 and 6.2.2.9).

• 2. This was fleshed out over the next few months in SKA Memo 130, a Preliminary System Description¹⁵³ for SKA Phase 1, based partly on the earlier High-Level Description document¹⁵⁴ (see Sect. 6.2.2.9). This document pulled together the array configurations (Fig. 6.7), the receptors, the signal chains, the signal processing (correlators) and the software required for imaging and pulsar observations in a single overview of the two telescopes envisaged in SKA Memo 125. It also provided tables and graphs of expected performance, which could be used by the science community to consider the impact on science of the phased approach. The emphasis was very much on SKA Phase 1 as a brief transit-point to the full SKA, and procedures were broadly outlined for how this could be done, while also incorporating the outputs of the Advanced Implementation Program into the full SKA.

• 3. In the meantime, an all-encompassing document, the Project Execution Plan,¹⁵⁵ (see Sect. 4.6.2) was being prepared. This was both an almost final, detailed report for PrepSKA and a plan for the next step, the pre-construction Phase of the SKA. At the system level, it covered management strategies, science drivers, system engineering, a system technical description, science operations and a description of work for pre-construction. It also covered options for a future governance structure, staffing, outreach, relations with industry (including intellectual property), and the contributions from the SKA pathfinder, precursor and design study projects.

  The breadth and thoroughness of this document was instrumental in obtaining the support needed to continue into the pre-construction phase.

• 4. For completeness, a delta CoDR was held in February 2011. This time it passed with only minor comments from the review panel¹⁵⁶.

At the system architecture level, this led in 2012/13 to a new baseline for SKA Phase 1,¹⁵⁷ a much more detailed description of the two telescopes, which by that time were to be established on two sites in Australia (SKA1-Low) and South Africa (SKA1-Mid). At that point the SKA project continued on a roll after the end of the PrepSKA era.

6.3 Historical Analysis of Individual Technology Innovations for the SKA

The purpose of the following sections is to describe the stories behind the development of essential parts of the SKA, concentrating on apertures (arrays of antennas) and on critical supporting technologies. Most of the development effort was devoted to innovative antenna designs, and the road to winnowing down selections was tortuous at times. Some designs were discarded by the developers, themselves, but others were selected by judgement calls using the advice of experts. In general terms, selections were guided by the science in the form of the Reference Science Mission¹⁵⁸ document. Cost, performance, and maturity played major roles. The title of this chapter, Innovation Meets Reality, illustrates the trend, especially towards 2012, the end of the PrepSKA program.

Critical supporting technologies were not subjected to the same rigour as apertures. It was recognised that many of these could not be fully selected until the aperture types were selected. Also, the rapid advances of optical and digital technologies meant that freezing selections too early would result in inferior designs.

6.4 Dishes for the SKA

The key to the success of the SKA at mid frequencies was clearly a new generation of dish designs. Traditional designs had evolved only slightly over the years. Until the Very Large Array (VLA) antennas were constructed, dishes were needed only in small numbers. While these dishes were optimised for the task, the structural design was fairly rudimentary.¹⁵⁹

Most importantly for the SKA, the cost of the dishes would be overwhelmingly the largest fraction of the budget. In 2007, a summary of dish costs, contained in SKA Memo 100 (see Sect. 6.2.1.4), indicated that they would be much more than half the total construction cost, even though their cost was underestimated at the time. This provided motivation to develop new techniques and designs for the SKA antennas and explains the dedication of time and resources during the PrepSKA period towards this end (see Sect. 6.4.5).

This led to a long period of innovation and development in two principal areas: structural design (Sects. 6.4.4 and 6.4.5) and sampling of the 3D space at the focus (two spatial and one frequency dimension) (Sect. 6.4.7). Although an enormous amount of progress was made and large resources poured in, none of these innovations made it into the final SKA telescope design. The risk was simply too high by the time the funding agencies were ready for the next phase and wanted a low-risk, secure design (see Sect. 4.6.2).

A reflector technology related to dishes is parabolic cylinders. Although abandoned as an option for mid and high frequencies in the early 2000s, they were studied intensively.¹⁶⁰ However, they later came back into their own in the CHIME telescope (Amiri et al., 2018) consisting of (at the time of writing) four 100 × 20-m cylinders at the Dominion Radio Astrophysical Observatory (DRAO) site in Canada. It has been extraordinarily successful at detecting fast radio bursts, an example of “exploration of the unknown” (see Sect. 6.2.2.8).

6.4.1 Dish Technology: A Thumbnail Sketch

Parabolic reflector antennas (‘dishes’) have been the mainstay of radio telescopes since pioneer Grote Reber built the first one in his back yard in 1937. Many examples of dishes are portrayed in Chap. 2. It might be assumed that their design has been refined over the following decades to the point where little more could be improved. Variations of reflector antennas including those with circular cross-section instead of parabolic (e.g., the Arecibo telescope), and parabolic cylinders (e.g., the Molonglo telescope, an SKA pathfinder ¹⁶²) have also been used for radio astronomy. ¹⁶¹.

Surprisingly, the antenna designs for the most recent array telescopes, the SKA and the next-generation VLA (ngVLA)¹⁶², have adopted designs that are unusual in radio astronomy. On the other hand, the designs for extremely large antennas used as single-dish radio telescopes have evolved separately and remain very different from those designed for arrays.

Among the thousands of articles on reflector antennas in the literature, the basics can be found most easily on Wikipedia. This is taken further in SKASUP6-7,¹⁶³ which discusses the operating principles of a radio reflector antenna. While dishes have a myriad of uses in communications, radar and in space, the radio astronomy receiving applications are by far the most demanding because the parameter space is large compared to more specialised applications.

In the early innovative period of the SKA developments, attempts were made to replace dishes at medium to high frequencies with other kinds of receptors, such as dense Aperture Arrays (Sect. 6.5.5) and Luneburg Lenses,¹⁶⁴ but the project always came back to dishes.

Later, several important attempts were made to radically reduce the cost of dishes while also improving performance. The story of why they were not adopted will be told here. Because the final design for the SKA dishes was adopted long after the time horizon of this book, this will be only briefly described.

6.4.2 Early Developments in SKA Dishes

In the early SKA context, the URSI Large Telescope Working Group (LTWG) discussed options for large reflector antennas (see previous section) in their first meeting.¹⁶⁵

The Allen Telescope Array,¹⁶⁶ originally called the 1-Hectare Telescope (1HT) and constructed at Hat Creek in California, was one of the earliest and most innovative of radio telescopes at the time and now (DeBoer et al., 2004; Welch et al., 2009). The first ATA Memo in 1998¹⁶⁷ captures the flavour of the discussions happening before it was first funded in 2001. It was the prototypical Large-Number—Small-Dish (LNSD) SKA architecture, championed in the US and adopted as a concept for the SKA. It bears a resemblance to the Radio Schmidt telescope that was proposed in the 1980s¹⁶⁸ but never constructed (see Sect. 2.2.2.4). The innovations of relevance to the SKA were: Continuous frequency coverage over a very wide band (0.5–11 GHz) using Wide-Band Single Pixel Feeds (WBSPFs); offset Gregorian dish optics; a novel mould-based antenna construction; and an 80 Kelvin cryogenically-cooled feed and low-noise amplifier. Although only the offset Gregorian optics design was adopted for the SKA in the end, the ATA provided design guidance with real system-level evidence in a way that other prototypical systems could not.

6.4.3 Dish Design Challenges in the SKA Context

Despite the simplicity of the basic reflector design, there are many design choices and parameters to be optimised. For SKA-Mid, a critical consideration is cost. As described in SKA Memo 100 in 2007 (see Sect. 6.2.1.4), the estimated number of 15-m dishes needed was between 500 and 600 for SKA Phase 1 and 2000–3000 for the full SKA (SKA Phase 2). The resulting A_e/T_sys figures of merit were 2000 m²/K for SKA Phase 1 and 12,000 m²/K for the full SKA. One could therefore simplify the goal to maximise A_e/T_sys per unit of currency for each antenna and hence also for the whole telescope array. In SKA Memo 100, the cost per 15-m dish was estimated to be about €300,000, including all components but not the associated infrastructure. Such mass production of radio astronomy reflector antennas had never been done or even contemplated before. Economies of scale were clearly part of the picture.

The following sections explain the key challenges and design choices that had to be made for the SKA dishes. Because there are many interlocking priorities, the design choices cannot be optimised separately; they must be optimised jointly, which leads to a compromise design for each of the important parameters. In subsequent sections, attempts to develop innovations that would break the traditional design-paradigm will be explained.

6.4.3.1 Diameter of the Dishes in the Architecture of the Telescope

One of the most contentious aspects of the SKA1-Mid design was the dish diameter. In the beginning both very large dishes (e.g., the LAR¹⁶⁹ and FAST^170,171 and very small dishes (e.g., ATA dishes (DeBoer et al., 2004; Welch et al., 2009)) were proposed. The dish diameter clearly determines the number of dishes needed to create a specified total collecting area. Note that both the effective area, A_e, and the field-of-view, Ω, are proportional to the square of the diameter, D²; for a given system temperature, T_sys. Both figures-of-merit (A_e/T_sys and survey speed (A_e/T_sys)² ∙ Ω) increase with diameter. The opposite driver is the cost-per-unit-area versus diameter.

Based on historical data, there were attempts to create cost versus diameter curves. However, it is exceedingly difficult to obtain sufficiently accurate cost information for antennas that were built over decades. The widely assumed curve relationship was C_D ∝ D^2.7, where C_D is the dish cost. If this were correct, the sensitivity per unit cost would decrease for increasing diameter for an individual dish, but this is not the full story for system cost.

The primary qualitative factors affecting the choice of diameter for a many-dish array of specified total collecting area are:

• 1 Large antennas can accommodate large feed packages, including cryogenic feeds which lowers T_sys dramatically. This is even more important if PAFs are included as options (see Sect. 6.4.7).

• 2 Small antennas require structural support pedestals and mounts for each one, which do not directly contribute to collecting area.

• 3 Small antennas have comparatively large fields-of-view, leading to higher survey speed.

• 4 Antenna maintenance cost is primarily per item. Many small antennas will increase maintenance costs proportionately.

• 5 Small antennas are easier to design for accurate pointing, surfaces, and alignment.

• 6 The signal and data-processing challenges for an array of many small antennas can be extreme¹⁷² (see Sect. 6.6.5).

Small dishes are still prized for their inherent field-of-view and ease of manufacture. For example, the Canadian Hydrogen Observatory and Radio-transient Detector (CHORD),¹⁷³ under construction in Canada at the time of writing, utilises technology developed for the SKA (see Sect. 6.4.4.2.1) for a 640-dish array (a 512-dish core and two 64-dish distant outriggers). This is a specialised telescope designed to ‘catch’ radio transients and to map the distribution of atomic hydrogen in the Universe over a large area of sky up to redshift 4.

6.4.3.2 Frequency Range

Another contentious aspect was the frequency range. The original motivations for the SKA were based on observations of the red-shifted spectral line of atomic hydrogen (HI-line) in galactic and extra-galactic radio sources (see Chap. 5), which is emitted by the source at a wavelength of 21 cm (corresponding to a rest frequency of 1420 MHz). Red-shifting, due to the recession velocity of the observed object, increases the wavelength of observation. As more astronomers became interested in the concept of a next-generation radio telescope, the science broadened, and initially the upper frequency was set to 3 GHz (λ = 10 cm) and later became 10 GHz (λ = 3 cm). Noting the rules of thumb outlined in SKASUP6-7,¹⁷⁴ the surface and pointing accuracies required for these frequencies are easily achieved by antenna fabrication methods.

However, projections for the SKA’s unparalleled sensitivity contained in SKA Memo 100 (see Sect. 6.2.1.4) brings another factor into play; wide-field imaging dynamic range.¹⁷⁵ Dynamic range refers to the capability of the telescope to image extremely weak objects in the presence of strong ones, rather like attempting to see a faint star near the Sun. Although this type of imaging is not the only scientific goal requiring high sensitivity, it is important to be able to detect and count radio sources emitting from the earliest stages of evolution of the Universe, which are inherently extremely weak.

For the SKA, the required dynamic range in the 15–30 cm wavelength range is approximately 10⁷, meaning that is possible to detect objects in the images that are 10 million times fainter than the strongest objects in the image. This will not be tested until the SKA is fully built.

Heuristically, it was argued that the shape of the antenna beam must be very stable to achieve sufficient dynamic range, translating directly into surface and pointing accuracy, even for long wavelengths. Tightening the rules-of-thumb by about an order of magnitude was required. Hence if that were the case, then extending the frequency range of the antennas would come along for the ride. A 10-GHz requirement remained the upper frequency limit until 2012/13; this governed the goals for the innovation programs described in Sect. 6.4.4.2.

20 GHz (λ = 1.5 cm) the final upper frequency requirement for SKA antennas was adopted in the SKA Baseline Design,¹⁷⁶ just at the end of the period covered by this book. In 2014–15, a survey of potential users of the SKA revealed that scientific interest in the highest frequency observing band (approximately 8–15 GHz) was ranked second behind the 21-cm wavelength band.^177,178

6.4.3.3 Noise

The introductory paragraphs to Sect. 6.4.3 explain the importance of limiting the value of T_sys, and as noted in SKASUP6-7,¹⁷⁹ the primary method of controlling amplifier noise is cryogenic cooling. Other sources of noise are spillover and scattering¹⁸⁰ (see Chapter 7). Early design considerations assumed that cryogenic cooling would be too expensive to operate because of the electrical power required and high maintenance costs, but a full cost-benefit analysis¹⁸¹ was not done until 2013/14, but by that time the benefits of cryo-cooling had informally been recognised.

Initially this left the antenna diameter unconstrained at the small end, although the ATA design (Welch et al., 2009), with 6-m dishes, used a clever combination of ultra-wide band feed with a cryogenic cooler to bring the amplifier temperature down to about 80 K, rather than the preferred 10–20 K. More generally, however, a set of cryogenic feeds of standard design establishes a soft lower limit to the antenna diameter in the 10-m range.

Spillover noise and noise from scattering are controlled mainly by the selection of the optical design of the antennas. This is considered in Sect. 6.4.3.4.

6.4.3.4 Optical Design

As one would expect, the choice of optical design, illustrated in SKASUP6-7, affects both performance and cost. For the ultra-sensitive SKA, apart from the diameter and frequency range, one aspect of the design stands out, scattering of the incoming radiation by the antenna structure. At the long-wavelength end, there will be components of the feed structure and focus assembly whose size is similar to a wavelength. This can result in resonant scattering, which is much stronger than normal at specific wavelengths. At the short wavelengths, scattering less important. Also, these two structural components block incoming radio waves. Both these effects subtly change the shape of the beam in ways that are difficult to predict and will certainly influence the imaging dynamic range. These considerations strongly argue for an offset design.

A major concern is the sensitivity of the antenna far away from the pointing direction, known as far-out sidelobes. RFI (see Sect. 6.2.2.12) can enter the signal path if it is not strongly rejected by the antenna beam. Scattering acts like an ‘omnidirectional antenna’; although its collecting area is small, RFI is extremely strong compared with natural signals from the sky. As the radio spectrum gets more and more crowded with man-made signals, the importance of reducing scattering increases. While not such a concern at the time when the SKA antenna design was being considered, new sources of RFI are now in the sky, itself, which increases the need to suppress signals far away from the pointing direction.

Measuring the polarisation of radio waves is a scientifically important aspect of radio astronomy. The polarisation performance of symmetrical dishes is usually considered better than off-axis dishes because the symmetry causes cancellation of undesired leakage between the two polarisation states of the incoming radiation. However, by structuring the optical design according to a method worked out by Mizusawa and Mizuguchi in the mid-1970s, an offset optical design with properties equivalent to a symmetrical paraboloid can be found.¹⁸² This approach was used in the SKA antenna design.

On the other hand, an offset design is structurally unbalanced and requires the sub-reflector and feed assembly to be supported by a cantilevered structure. This is significant because, ignoring its self-weight, the deflection of a cantilevered beam loaded at one end scales as the cube of its length (L³).

Another cost-driven consideration of an offset design is that the primary reflector has an elliptical outline. But the projection of the reflector in the pointing direction is circular. Hence, more physical area must be built than can be used as collecting area, adding to the cost. For example, the primary reflector of the SKA dishes is 15 × 18 m, but the projected collecting area is only a 15-m diameter circle. Moreover, the traditional method of building the reflector is to tesselate the surface with triangular or four-sided panels. A symmetric design has fewer different panel shapes than the elliptical reflector in an offset design; therefore, manufacturing cost of a design that uses panels is increased.

In summary, the structure of an offset antenna is significantly more expensive than a symmetrical antenna for the same collecting area. But the advantages of the offset design were considered so significant that they became a requirement for the ultra-sensitive SKA to be built. This is especially true for the 10–50 cm wavelength range. Hence the offset design was selected, and for additional technical reasons, the offset Gregorian optics was decided upon (see Sect. 6.4.5).

SKASUP6-8¹⁸³ contains a summary of the desirable properties ultimately agreed upon for the SKA.

6.4.4 Ambitious Innovations in Antenna Structural Design

The traditional structural form for the antenna consists of a small number of components (see the Box in SKASUP6-7¹⁸⁴): a pedestal, a mount (turnhead) which contains the axes of motion and motors, a back-up structure which supports the panels, and feed arms which support the sub-reflector and feed assembly.

Because they are large structures, often in an open environment,¹⁸⁵ dishes are subject to so-called load cases, the primary mechanisms of structural distortion. These are gravity, wind, solar illumination, and temperature. Less important are rain, humidity, flooding, etc. The timescales on which the load cases act are important: wind can act on time-scales of 10s of seconds to minutes, whereas solar illumination is minutes to hours, and temperatures are longer. Of course, gravity is static, but its influence on the structure changes with pointing direction. All greatly affect and even drive the antenna design.

The following sections and SKASUP6-9¹⁸⁶ describe numerous attempts to break this design paradigm in a drive to both reduce the cost of antennas and to adapt to the requirements traceable to SKA science goals. A helpful aspect was the number of large antennas to be produced, an opportunity that had never previously arisen. Initially the full SKA would have contained about 3000 antennas, but this was later whittled down to 200 in the first phase of construction. Nevertheless, the potential savings from quasi-mass-production promised to be significant if the designers were clever enough to take advantage of it. Cost reduction is best measured in terms cost per unit of sensitivity (€ per unit of A_e/T_sys or € per unit of survey speed).

6.4.4.1 Sky-Mount Antennas

The emergence of ‘sky-mount antennas’, a term coined in Australia, was strongly coupled with the development of phased array feeds (PAFs) (see Sect. 4.3.3.1). These feeds are a compact array of small antennas which are combined in a way that is similar to an aperture array (see Fig. 6.19 and the box in SKASUP6-11¹⁸⁷). While the method is based on sampling the electromagnetic field around the focus, the effect is to produce a grid of beams on the sky in the place of a single beam, hence increasing the field-of-view by a factor of the number of beams (typically 36). This dramatically increases the survey speed of the telescope if the A_e/T_sys component of survey speed can be maintained (Sect. 6.4.3.1). But to work properly for an aperture synthesis telescope, the pattern of beams must be fixed to the sky. As explained in Sect. 6.4.5, this was done by adding a third axis, orthogonal to the reflector, to counteract the rotation of the reflector against the sky (see Fig. 6.10). CSIRO, the lead radio astronomy institution in Australia, was keen to develop these ideas for the SKA.

Fig. 6.10

The ASKAP antennas under construction. The third axis of rotation is located just under the reflector. Credit: CSIRO Radio Astronomy Image Archive CRAIA-SKA011

By 2009, ASKAP had become a major facility project (DeBoer, 2009) carrying a large scientific program, well beyond the concept of merely testing PAFs. By this time CSIRO had a great deal of confidence in PAFs and how to build the supporting infrastructure, such as beamformers, for a reasonable cost. More detail on ASKAP as a whole is contained in Sect. 4.3.3.1.

6.4.4.2 Mould-Based Reflectors

The most ambitious innovations were to fabricate the entire primary reflector in one piece by forming it over a mould. This approach began in three different places, probably independently: Canada, USA, and South Africa. Although the approaches differ in detail, they followed a common theme: a convex mould was constructed over which the reflector material was formed by compressing the material against the mould. The concave reflector part was released from the mould as a complete reflector.

There were several potential attractions to this approach:

• Rapid fabrication in mass production: Although some methods require a cure time of a few hours, the complete reflector is available after it is released from the mould.

• Repeatability: With adequate control over the fabrication environment, the reflectors will be nearly identical. No adjustments are necessary.

• Accuracy: The accuracy of the reflector is determined by the accuracy of the mould. Very accurate reflectors are possible.

• Variety of shapes: Asymmetric shapes or bi-lateral reflector symmetries (or any dual-curvature shape) can be fabricated as easily as symmetric shapes. In contrast, panelled reflectors are much easier to make for symmetric reflectors, but more expensive for offset reflectors.

• Partly Self Supporting: Because of its continuous surface, the reflector can be designed to carry most of its own weight with much less underlying structure than panelled reflectors, especially if it is highly curved.

Mould-based fabrication of large reflectors for the huge number of radio telescopes being planned (for SKA Phase 1 and the much larger SKA Phase 2), promised an enormous impact on cost and feasibility. While retaining the tried-and-true advantages of reflectors, comparatively frequency-independent and passive collecting area, mould-based reflectors might have revolutionised the field. Sadly, it was not to be.

In Canada and South Africa, fibre reinforced plastic (carbon fibre and fibreglass, respectively) was used to build prototypes. In the USA, hydro-formed aluminium was used for the ATA antennas (for examples of mould-based reflectors, see Figs. 6.11, 6.12, 6.13 and 6.14). These efforts are described in more detail in SKASUP6-9.¹⁸⁸

Fig. 6.11

Left: Vacuum Infusion of the 10-m Mk2 reflector at the Dominion Radio Astrophysical Observatory in Canada on June 4, 2008. Credit: National Research Council of Canada. Right: The reflector with the (PHAD) feed completed in October 2008. Credit: The National Research Council of Canada

Fig. 6.12

Left: The XDM telescope mould under construction at Hartebeesthoek Radio Astronomy Observatory in South Africa in 2007. Right: The XDM with all subsystems installed, integrated and tested in 2009. Credit: South African Radio Astronomy Observatory (SARAO)

Fig. 6.13

The KAT-7 fibreglass dishes under construction in the Karoo district of South Africa in 2009. Right: The completed KAT-7 dishes being installed in 2009 on what is now the site for MeerKAT and SKA1-Mid

Fig. 6.14

Left: The first ATA reflector on the factory floor. Right: A completed antenna. Credit: David DeBoer

6.4.4.2.1 Mould-Based Reflectors in Canada: Carbon Fibre Reinforced Plastic (CFRP)

The Canadian project began in early 2006 as an outgrowth of the Large Adaptive Reflector (LAR) project,¹⁸⁹ when the National Research Council of Canada hired Gordon Lacy in 2004 to work at DRAO. With a background in composite fibre construction techniques in the marine industry,¹⁹⁰ he was hired to design a light-weight support for the suspended LAR feed, which was based on PAF technology (see Sect. 6.4.7.1.1). The feed support was to be made from Carbon Fibre Reinforced Plastic (CFRP).

The project began with the production of a mere 1-m offset reflector. It progressed to two 10-m symmetrical versions, the second of which displayed a root mean square (rms) surface accuracy of 500 microns,¹⁹¹ sufficient for high efficiency at 20 GHz, confirmed with a laser tracker and holography in October, 2008 (Fig. 6.11). The next in the series was Dish Verification Antenna 1 (DVA-1), a prototype designed to the SKA specifications at the time. This played a major role in the history of SKA dish design for several years, as told in Sect. 6.4.5.

6.4.4.2.2 Mould-Based Reflectors in South Africa: Fibreglass Reinforced Plastic

Planning in South Africa began in early 2006 for the Karoo Array telescope (KAT), the first of which was a 15-m diameter prototype (XDM) erected at the Hartebeesthoek Radio Astronomy Observatory (HartRAO) (Fig. 6.12). The XDM prototype was part of a larger plan funded by the South African government for up to US$50 M, among other things to build and equip the dish with state-of-the-art receivers and digital back-end devices. Plans were also laid to build 20 dishes on the Karoo site in the Northern Cape region, which had been selected for the South African bid for the entire SKA.¹⁹² This was later trimmed to 7 dishes, becoming the KAT-7 project (Fig. 6.13).

The KAT-7 array was commissioned by the end of 2010, but by that time the SKA had adopted 15-m offset reflectors as the SKA standard. Also by that time, it had become clear that MeerKAT antennas were unlikely to use the simple mould-based method developed for the XDM and KAT-7, which was restricted to building only symmetric antennas.

6.4.4.2.3 Hydroformed Aluminium Reflectors in the United States

As outlined in Sect. 6.4.5, the TDP program was funded in late 2007 (see Sect. 6.2.1.3) and had assembled an Antenna Working Group (AWG) to globally coordinate dish development. They also planned to carry out innovative dish development themselves, concentrating on the hydroforming process.

The ATA (DeBoer et al., 2004) was the first radio astronomy telescope that had used a hydroforming process for fabricating its 6-m diameter reflectors (Fig. 6.14). Little has been published about the proprietary process¹⁹³ used to make the ATA reflectors or the so-called Deep Space Network Breadboard reflectors.^194,195 The process was similar to one widely used in industry (Leuthesser & Fox, 1955) for making small reflectors for satellite television reception, sheet-metal parts for automobiles, etc. The method employs a high-pressure fluid (not necessarily water as implied by the name) to deform a flat metal sheet over a mould, resulting in a sheet that has taken the shape of the mould.¹⁹⁶

In 2004, prior to TDP funding, the US SKA community proposed a symmetrical 12-m hydroformed design with a 2-m mesh extension (Schultz et al., 2004) which would yield a 16-m antenna. By 2009, this proposal had evolved to 12-m hydroformed antennas with either symmetric or offset Gregorian optical designs¹⁹⁷ without extensions. This was smaller than the 15-m antennas adopted by the international project as the optimum size for minimum overall system costs.

A major impediment to further development of the hydroforming approach for the SKA dishes was the up-front capital cost¹⁹⁸ required to carry out a full investigation of a 12-m hydroformed design and to build a prototype (see point {C} in Chap. 6 introduction). It was this cost and the limitations to the maximum antenna diameter possible with hydroforming that led to it being discontinued for the SKA. Instead, the TDP moved to joint development of mould-based 15-m CFRP antennas together with the Canadians and other partners around the world (see Sect. 6.4.5.1).

6.4.4.3 Preloaded Parabolic Dish (PPD): Tension Structure Reflectors

First proposed for the SKA in 2002,¹⁹⁹ the Raman Research Institute began an experiment to construct a PPD dish in 2003 using the principles of preloaded structures, based on a design from the Tata Institute of Fundamental Research in Pune, India.²⁰⁰ As they described it (Shankar, 2008), “The preloaded concept is based on the principle that if a structure has an initial stored strain energy, then under certain conditions, it has the capacity to offer a larger stiffness for the same weight to additional external loads.”.

A 12-m prototype PPD dish was completed in 2008 and measured photogrammetrically,²⁰¹ yielding rms deviations from an ideal surface of about 20 mm (Fig. 6.15). Because these dishes were suitable only for low frequencies, it became clear that they could not be adopted for SKA-Mid, especially after preliminary specifications were published in December 2007 (SKA Memo 100, see Sect. 6.2.1.4). And for low frequencies, more flexible aperture arrays were a better match to requirements over dishes.

Fig. 6.15

The fully equipped PPD antenna at the Gauribidanur Radio Observatory. Credit: The Raman Research Institute

6.4.5 Historical Twists and Turns in Dish Development

As previously noted in the introduction to the section on SKA dishes (Sect. 6.4), the cost of the dishes would be overwhelmingly the largest fraction of the budget. Beginning in late 2007, the basic specifications for SKA dish antennas had been roughly outlined in SKA Memo 100. Almost all the detailed parameter space was still open, but the expectations were clear:

• The frequency range for dishes, initially based on the science, was also influenced by the technology it was paired with. If paired with dense Aperture Arrays covering frequencies from 0.5 to 1.5 GHz, dishes needed only to cover frequencies above 1.5 GHz. Otherwise, the lower limit for dishes would be between 0.3 and 0.5 GHz. The upper limit considered was either 3 or 10 GHz, depending on cost.

• Three types of feeds were discussed: (1) Wide-band single-pixel feeds (WBSPFs, see Sect. 6.6.1.1), which cover a frequency range from 0.5 to 10 GHz (20:1 range), were assumed to be on the near-term horizon for development. (2) Traditional feed systems consisting of several switchable feeds with only a 2:1 frequency range were also a possibility, but cost and feasibility were clearly a problem, especially for small dishes. (3) PAFs (see Sect. 6.4.7), which are arrays of small antennas designed to produce multiple beams on the sky. PAFs greatly expand the instantaneous field-of-view of a single antenna, thus ‘reusing’ the collecting area of the antennas, but the beam-pattern had to be stabilised on the sky (see Sect. 6.4.4.1).

• Cryo-cooling of the first stage of amplification was widely used in dish antennas. Innovative techniques for single-pixel feeds were assumed to be possible using Stirling cycle devices, a cost-effective compromise able to cool to 80 Kelvin instead of the usual 20 K. There was also a hint that uncooled amplifiers might be developed that would not require cryo-cooling. However, for PAFs, which had to be very large at mid-frequencies, the possibility of cryo-cooled amplifiers was considered remote.

None of the SKA Memo 100 specifications stuck in the end, but the boundaries of parameter-space were clear, and under the aegis of PrepSKA, its exploration was quickly systematised, beginning in 2008. Above all, it was also clear that priority had to be given to reducing the cost of antennas.

In November 2008, a plan^202,203 was formulated to systematise all PrepSKA work packages: organise deliverables, assign roles and responsibilities, and to conform to system engineering principles (see Sect. 6.2.2.2). A parallel document,²⁰⁴ setting out the rationale for dish development, specifically covered dishes in the context of an integrated approach to the entire SKA telescope system.

In the rationale document, four model development programs were defined: (1) An “optimised ‘single-pixel feed only’ antenna” to deliver the baseline described in SKA Memo 100 using optimised but well understood technology. This was the design against which the other three designs, containing more advanced technology, were to be measured. The other models were essentially optimised differently and designed to handle both WBSPFs and PAFs. (2) An Alt-Az Mounted Antenna with on-axis feeds to provide the lowest cost model that could accommodate both types of feeds. (3) The sky-mount design (see Sect. 6.4.4.1) to provide a model that did not emphasise the clean-beam design (i.e., normal scattering), but did emphasise freezing the parallactic rotation²⁰⁵ of beam patterns on the sky. The third axis was introduced to produce counter-rotation, thus freezing the antenna pattern on the sky. (4) An offset optics antenna to provide the best scattering performance.

Each development was to be led by a ‘lead institution’, but contributions from all the SKA participating institutions were to be made to all four models in a kind of matrix organisation. Standard system engineering processes were invoked. A Work Breakdown Structure (WBS) and a schedule were developed, ending in June 2010. Based on the design work, the goal was to eliminate two of the models and to build jointly funded prototypes of the remaining two.

In fact, no lead institutions were formally appointed. In the case of dishes, a TDP report to the SKA Science and Engineering Committee (SSEC) in February 2009²⁰⁶ contained a statement: “Decision that the TDP will serve as a clearinghouse for antenna work done around the world, consolidating information for use by the international SKA project. Close collaboration between TDP and DRAO on composite reflectors, mounts, and phased array feeds. DRAO participation in TDP industrial studies.”. This was accepted by the SSEC.

The TDP had already formed an Antenna Working Group (AWG) (see Sect. 6.2.1.3). The first meeting was held in San Francisco in March 2008 at which the PrepSKA work packages were described in detail²⁰⁷ by Peter Dewdney, the International Project Engineer: overview of SKA science goals and top-level system specifications; review of proposed stages of build out; cost drivers; and detailed discussion of reflector antenna specifications. This was followed by detailed discussion and negotiation of the design processes at the TDP, mindful of their roles as both clearing house and proponent of a specific design.^208,209 Participants from the ATA were especially Influential in this group and strong proponents of very small dishes.

In an important side-meeting at the San Francisco event, top-level specifications for SKA dishes were worked out between the SPDO, DRAO and TDP representatives.²¹⁰ This eventually became the accepted basic design for the SKA-Mid antennas, although it took a long time to be fully accepted.

6.4.5.1 The Saga of DVA-1: The SKA Prototype Dish

The Large-Number—Small-Dish (LNSD) antenna array concepts had been defined as part of the SKA Reference Design in early 2006. In the 2 years following, pressure mounted to define the actual dish design, and nail down what was clearly the largest single cost-component of the SKA. A high specification, low-cost design was needed if the SKA were to come even close to meeting the projected cost targets. As described above, it was considered that traditional designs could not deliver this combination, and so a great deal was riding on the mooted potential innovation aspects and on the decision-making process.

As described above, beginning in 2008 the TDP’s task was to act as a clearing house for antenna design, in addition to their own program, which called for the development and production of an SKA prototype antenna. The broad outline was discussed at the San Francisco meeting in March, 2008.^211,212 Most of the TDP effort between mid-2008 and 2009 was devoted to (1) dish optics studies^213,214,215 (2) the design^216,217,218 and (3) performance and cost.²¹⁹ An offset-Gregorian optics design clearly had had momentum for a long time, but questions of the diameter, implementation and cost remained. As described in Sect. 6.4.3.4, the cost of an offset antenna was higher than a symmetric design, but not so much that the difference could not be afforded, especially considering the higher A_e/T_sys performance metric for offset antennas. In parallel with work on other aspects, work on comparing dish costs continued to the end of 2010.²²⁰

Meanwhile, the SPDO was focussed mainly on bringing the SKA dishes into a system engineering framework (see Sect. 6.2.2.2) as part of the larger SKA-Mid design.²²¹ The resulting analysis was described in an SPDO Memo²²² and in documents that defined a Dish Verification Program.^223,224

In February 2009, the SPDO drafted a TDP/DRAO/SPDO agreement²²⁵ to (lightly edited) “carry out a joint program of work on PrepSKA WP2 Tasks according to an agreed timescale and with agreed deliverables and follow an agreed project management methodology—as outlined in the Guiding Principles document” (see Sect. 6.2.2.2). It is not certain that this agreement was ever signed, but it did provide a mutual understanding of the roles and responsibilities of the parties, a detailed work-breakdown structure, and a timeline.

The question of the diameter (see Sect. 6.4.3.1) had been unresolved for some time and a solution had to be found to make progress with a prototype, although even this was questioned by the TDP which argued that several prototypes would be required and the diameter of the first prototype would not need to be final. The TDP favoured a 12-m diameter, mainly because they probably would not be able to carry forward the hydroforming option for a 15-m antenna. Also, they were not really making progress on a 12-m hydroformed antenna, mainly due to the huge cost of tooling.

The SPDO favoured a 15-m antenna, based on an analysis of capital and operational costs (e.g., maintenance) for a smaller number of antennas.²²⁶ This analysis was validated separately in a discussion document on choosing 12-m or 15-m dishes.²²⁷ Nevertheless, it was not absolutely clear to anyone that any 15-m mould-based antenna could be prototyped, but it seemed probable that a mould-based composite carbon fibre reflector (see Sect. 6.4.4.2.1) could be built. As a fall-back, a traditional panelled dish would be pursued.

The positions of the TDP and the SPDO were made clear in an interchange of notes²²⁸ in which the SPDO outlined their key requirement: the deliverable had to be a “near-production prototype dish that had been carefully tested”. The TDP was inclined to view the dish verification program as consisting of a “first prototype of a small converging series of prototypes”, DVA-1, DVA-2, possibly with different diameters. Also, the SPDO emphasised that the diameter should be based on system cost and performance, not on the cost of a prototype. The TDP was very concerned that abandoning hydroforming would undercut their credibility with the US National Science Foundation, which was funding them.

The over-riding consideration was the timescale. The PrepSKA program had to demonstrate a costed, tested dish design. By the end of 2009, the TDP and other proponents of dish designs realised that much closer collaboration was necessary. The TDP put forward a draft Dish Verification Plan,²²⁹ which identified potential partners including NRC/DRAO which could provide reflectors, NRAO which could provide a test site, CETC54 in China which could provide entire antennas, and the South African SKA Project Office which was interested in collaborating on antennas. It was emphasised that the selection of the basic design parameters would need to be made early in 2010.

In the meantime, the South African group approached NRAO with a collaboration proposal involving NRC/DRAO for which NRAO would provide design, testing expertise and a test site, while NRC/DRAO would provide its mould and composite dishes. However, NRAO expressed a general concern that the whole dish development effort would be fragmented. In fact, fragmentation did happen briefly but shortly afterwards, a group consisting of the SPDO, TDP and DRAO coalesced without NRAO and South Africa. Tentatively, they agreed to build a 15-m prototype if the funds could be found, otherwise a 12-m prototype⁶²⁸.

A major DVA-1 meeting was held in Arlington, Virginia in April 2010 to discuss all aspects and alternatives for the DVA-1 design, including alternatives from potential suppliers, such as CETC54. The result was not definitive, but a smaller group led by the TDP arranged a side meeting at which a more detailed description, “Selected Baseline Design”, was hammered out.²³⁰ This was a 15 × 18-m, shaped²³¹ offset design with a 4-m sub-reflector with space at the focus for a variety of feeds, including WBSPFs. What remained to decide is whether the feed should be mounted above the reflector (‘feed-up’) or below (‘feed-down’).

Although at arm’s length, since it did not control the national funding involved, the SPDO continued to stress the need for system engineering rigour and produced a top-level specifications document for DVA-1²³² which specified the size and optical configuration but was agnostic on the fabrication method. This had to be kept general because there were still competing visions for the SKA dishes. A Dish Conceptual Design Review (CoDR) was held in Socorro in February 2011, with review-panel members that were acknowledged experts in antenna design. The project plan presented at this CoDR consisted of the fabrication of a 15-m composite reflector by the DRAO team, to be built on the ALMA antenna test site at the NRAO VLA site in New Mexico (see below). Although the panel expressed concerns over the technical risk of the rim-supported composite reflector and a lack of a clear management structure, it concluded that the project was ready to proceed to a more detailed, preliminary design stage.

The SPDO and others were concerned about qualifying and testing the prototype if it were just a single dish at the DRAO site. A test plan involving the construction of DVA-1 (or a copy) on the VLA site had been put forward in 2009 by the SPDO.²³³ This would involve a battery of single-dish tests, followed by tests using the Expanded -VLA (EVLA) antennas in interferometer mode with the prototype. Most of the proposed tests would require only a few VLA antennas, but one or two tests would require the entire array for a short time. Several people in the US questioned whether the interferometry segment of the plan would work. Nevertheless, it was eventually agreed that even single-dish testing at the EVLA site would be very useful and bringing in expertise from NRAO would be important.

During mid-2011, a Letter of Intent was being circulated that established a framework for the design, construction and testing of DVA-1. After delays resulting from legal concerns, it was signed by Cornell University, the Herzberg Institute of Astrophysics (NRC in Canada), NRAO, and the University of Manchester (representing the TDP, DRAO, NRAO and the SPDO, respectively). This group formed a Management Board for the DVA-1 project, chaired by Bob Dickman of NRAO. An engineering team led by a Project Engineer from the University of California (Berkeley), a Project Manager from NRAO, and technical experts were assembled using a combination of funds and contributed personnel.²³⁴ Their near-term goal was to bring the project to meet the criteria for a Dish Subsystem Preliminary Design Review.

By September 2011, the TDP was at the end of its term and needed an extension of funding to continue. The US National Science Foundation (NSF)) intended to hold a “Programmatic Review” to assess the technical progress of the project at the CDR level as well as the value of the DVA-1 program to US radio astronomy beyond the SKA. However, in the light of the ASTRO2010 decadal report (see Sect. 4.5.3), the NSF was reluctant to continue funding all aspects of the TDP program but was willing to consider the DVA-1 development separately from the SKA on the basis that an innovative dish program might be useful in a future project. Nevertheless, they did permit the TDP to use its remaining resources to support the DVA-1 program. However, differing goals among the partners led to tensions within the project.

A Preliminary Design Review (PDR) was held for DVA-1 in October 2011; the review panel recommended that the project proceed to final design and hold a Critical Design Review (CDR).²³⁵ However, by early 2012 the TDP funds had dried up and the project had to consider a ‘Plan B’. The Canadian NRC agreed to carry on, provided there was support from PrepSKA for the build phase, especially the pedestal and mount, which was close to production-ready in the US.

The DVA-1 project was now ‘under one roof’, in Canada, with one goal, to provide the SKA with a qualified dish, as described in (Lacy et al., 2012).

A CDR was held in July 2012 and construction began with a combination of NRC and PrepSKA funds.²³⁶ This allowed the DVA-1 pedestal and mount to be built and sent to the DRAO site and provided support for key TDP personnel to continue.

The CDR review was carried out by five internationally recognised experts in the field. Their report²³⁷ was very complimentary, while noting a few reservations on the maturity of some aspects of the design, understandable given the number of design innovations.²³⁸ In particular, they noted: “The key aspects of the DVA-1 design include:

• 1. Use of a single-piece rim-supported carbon fibre reflecting dish that serves not only as the primary optical element but also as a major structural component,

• 2. Use of an unblocked offset optical design leading to a very clean antenna pattern,

• 3. Use of shaped (non-conic section) reflectors that are used to maximise the antenna gain while still minimising side-lobe response,

• 4. A secondary focus point optimised for today’s leading wide-band feeds,

• 5. Clear access to both the secondary focus and primary focal region.

While DVA-1 was not the first antenna to demonstrate these individual aspects, it was the first to apply all of them. Moreover, they were being applied to an impressively large 15-m by 18-m structure with surface tolerance requirements of 1 mm or better, while encountering a wide range of environmental conditions. Collectively these features offered promise of high antenna performance at low cost.”

In addition to these innovations, the design team also demonstrated a reflectivity of the surface comparable to that of aluminium while maintaining a protective dielectric cover over the reflecting surface. The DVA-1 design also employed a semi-compliant diaphragm at the centre of the main reflector which reduced the need for a backup structure while maintaining sufficient stiffness against transverse wind loading. Moreover, the DVA-1 mount was also an innovative design. Its innovative features are neatly summarised in one of the DVA-1 CDR documents.²³⁹ The key features of the design were later documented in more detail in an NRC report.²⁴⁰ It is unfortunate that this group of features have yet to be found in a production antenna design.

Overall, this gave the SPDO reason to support this work in the expectation that a cost-effective, high-performance design would emerge.

Nevertheless, in the months leading up to the DVA-1 CDR, the SPDO was concerned about coverage of all the aspects of dish design for the SKA in reviews, eventually resulting in a tree-structured mind map.²⁴¹ The number of items beyond just an innovative and cost-effective structure indicated that it would be impossible to cover them all in a single review. This led to a persistent concern in the SPDO.

At this point the option for testing the DVA-1 antenna on the VLA site in interferometric mode disappeared. The level of funding required from Canada to build a facility there was more than was available. This highlighted one of the principal shortcomings of the composite design, the one-piece reflector required fabrication on site. The DRAO group carried out several studies on building portable fabrication facilities but was not able to follow through, and the alternative of fabricating multi-piece reflectors was thought to have significant cost and performance disadvantages.

At the same time DRAO found that they did not have the space to comfortably construct a 15 × 18-m reflector on the DRAO site and decided to build it in a nearby town which had a suitably large building. After working out weights, they realised that a large helicopter would be capable of lifting the main reflector and carrying it over the intervening mountain to the site. Moulds for the two reflectors were assembled there and construction began (Fig. 6.16 top). Several improvements to the design were also incorporated.

Fig. 6.16

Top Left: The mould for the DVA-1 main reflector (16 × 18 m) under construction. Top Right: The DVA-1 mould for the sub-reflector (4 m diameter). Bottom left: The DVA-1 reflector unexpectedly becoming airborne and being abruptly halted by the rigging during helicopter transport. Credit: Gary Hovey. Bottom right: The reflector after being “popped back into place” and repaired. Credit: The National Research Council of Canada

In Oct. 2013, DRAO was ready to fly the reflector to the site. The dish was suspended under the helicopter by a long tether (Fig. 6.17). The first part of the journey went just as planned, but on the final approach the helicopter decelerated faster than the reflector causing the dish to swing ahead of the tether point. It then became airborne and when the reflector fell back down, the rigging halted its fall abruptly. The force from the deceleration caused the concave reflector to invert to a crumpled convex shape, apparently destroying a now inverted surface.

Fig. 6.17

The DVA-1 15 × 18-m reflector being flown from the fabrication site near Penticton, B.C. to the DRAO site. Credit: The National Research Council of Canada

It seemed very unlikely that the reflector could be recovered. However, the head engineer suspected that industrial airbags used to right over-turned trucks could be used to ‘pop’ the surface back to its original shape. In an extraordinary effort, the team managed to successfully carry this out (Fig. 6.16 bottom). Although a few surface repairs were needed, the accuracy of the surface was retained to the original specification except for one small area. This was certainly a demonstration of the incredible resilience of the carbon fibre surface design.

While this was seen as a disaster at the time, it did demonstrate that a reflector suitable for the SKA could be transported from a central fabrication site to where it was to be installed. The fact that a helicopter can easily carry the weight illustrates an important advantage of CFRP construction. Of course, several more precautions would have been needed, such as providing some reinforcement of the surface during transport. Figure 6.18 shows the result after repair. Amazingly, the performance met design specifications, which at the time were for high efficiency at 10 GHz.

Fig. 6.18

The DVA-1 dish, after repairs and shortly after the sub-reflector was mounted. Credit: The National Research Council of Canada

A significant design aspect was whether the feed should be located above the reflector (“feed high”) or below (“feed low”). There were, and still are, pros and cons, and this was debated extensively. The structural issues were covered in a document presented at the DVA-1 CDR (June, 2012).²⁴² Structurally, the feed-high design presented advantages: the structure could be almost balanced on the pedestal and the support structure behind the reflector could be stiffer, leading to a compact design. The problem with the feed-low option was mechanical interference with the pedestal when pointing close to the horizon; in that case the elevation axis had to be cantilevered out from the top of the pedestal, or the back-up structure had to be split to avoid hitting the pedestal. In the analysis of a TDP Antenna Design Note,²⁴³ this led to at least a 30% increase in capital cost. On the other hand, access to a high feed would require an elevated platform; the implications of this would certainly increase operations cost, but a thorough study of this was never carried out.

A third consideration was the impact on system noise originating from feed spill-over, radiation entering the feed from the ground (see Sects. 6.4.3.3 and 6.4.3.4). A thorough analysis²⁴⁴ did indeed see this effect but concluded that only a modest increase in noise at low elevation angles would be incurred. In any case, radio astronomy observations attempt to avoid these angles because of increased atmospheric noise. However, this document did also note that for the feed-low option it was possible to shield the feed from the spill-over noise, but not possible for the feed-high option. The final call was feed high.

The feed-high/feed-low discussion became a controversy, and eventually came back to haunt the DVA-1 project, mainly because of the operations cost, but also because of the ability to shield spill-over noise. Ultimately, the feed-low option was chosen for the SKA. This forced the NRC/DRAO to re-design the antenna as they readied themselves to build a second prototype. The original DVA-1 antenna is still in use for observations at 20-cm wavelengths but has never been fully tested to SKA specifications.

Although slightly beyond the scope of this book, an SKA Dish Consortium (SKADC) had been formed with the responsibility for choosing a dish design for the SKA. By that time three other concepts were being proposed, including the Single-piece Rim-supported Composite (DVA-1/SRC). The SKA required a low-risk technology encompassing as little development as possible.²⁴⁵ A dish with panels, rather than a single-piece reflector, supported by a traditional space-frame backup structure was selected. The overall cost of SKA-Mid led to a descoping of the project, including a reduction of the number of new antennas for SKA1-Mid while making up the difference by incorporating the new MeerKAT antennas for lower frequencies.

The DVA-1 saga came to an end, at least for the SKA. The planned DVA-2, designed to meet the 20-GHz specification, was eventually built. Nevertheless, development of technology continued at a low level, with designs proposed for the next-generation VLA (ngVLA) ²⁴⁶ antennas. As well, small-scale commercialisation of the technology occurred for satellite communications at high frequencies.

Looking back at this story, it is easy to see reasons why the original intentions for innovation were not finally realised:

• Many relatively new technologies had to be incorporated into the design.

• Traditional structural designers and fabricators could not take on board the use of carbon fibre composites, which require a completely different knowledge base.

• The National Research Council of Canada (NRC Canada) devoted significant resources over the years, but still not enough to ‘put it over the top’. In this respect, it is like the hydroforming program in the United States, although the effort in Canada was many times larger.

• Apart from the fading US contribution, the SKA community, itself, was not entirely behind the DVA-1 project, making international collaboration difficult. Other countries were still promoting options to either replace dishes entirely with aperture arrays, or were interested in other aspects of dish design (e.g. PAFs -  see Sects. 6.4.4.1 and 4.3.3.1).

• The commercial market for large antennas is small. A large market has been the key to the adoption of carbon composites to other large structures, such as for aircraft. NRC Canada would have been much more interested and helpful in partnering with commercial organisations if significant commercial opportunities had been available.

• A way to tunnel through capital and collaboration barriers was needed. Perhaps only huge companies not beholden to taxpayers can afford this kind of risk. And perhaps they are huge because enough technology bets have paid off.

It is impossible to tell whether a real opportunity has been lost, or whether this technology would never have been competitive anyway. However, more persistence, time and resources would have made a big difference. If the US had not dropped out of the SKA, it would have been much more likely that resources and improvements in the practicality of the design would be found to bring the DVA-1 technology to maturity (see point {C} in Chap. 6 introduction). Nevertheless, as noted in Sect. 6.4.3.1, the CHORD project has selected technology developed for DVA-1 for simpler, smaller dishes. This may be a steppingstone to the future.

6.4.5.2 Independent Dish Array Developments, Precursor Telescopes and Impasses Leading to the System Concept Design Review

As it turned out, the dish development process was not as clean or clear-cut as initially planned in the Guiding Principles document (see Program 4 (P4) in SKASUP6-1²⁴⁷) and lack of focus was also noticed at the System CoDR in 2010 (see Sect. 6.2.2.9). National priorities took over, although cross fertilisation continued and there were many cooperative meetings over the next few years, led by the SPDO with the larger project in view. CSIRO persisted in developing a sky-mount design (see Sect. 6.4.4.1), eventually leading to ASKAP. SKA South Africa, as it was then known, continued with their own design iterations, ending with a version of an offset reflector, that led to MeerKAT antennas. ASTRON funded the design of a small, prime focus antenna that would fit with their interest in mid-frequency Dense Aperture Arrays as part of the SKADS Benchmark Scenario (see Sect. 6.5.5.2). In India a design for a reflector based on tension structures was being designed and was later prototyped (see Sect. 6.4.4.3). The TDP initially pursued hydroforming technology for fabricating reflectors, as used for the ATA dishes (see Sect. 6.4.4.2.3), but later collaborated with NRC/DRAO in Canada to build a prototype offset reflector using carbon fibre fabrication technology.

6.4.6 Cost Estimates for Dishes

The sensitivity of the system design to dish costs and the number of dishes was well understood.²⁴⁸ Attempts were made to estimate the costs of dishes throughout the period covered in this book as proponents of the different dish designs were searching for ways to optimise the SKA system, especially to model the scaling of dish diameters and frequency ranges with cost.²⁴⁹ Historical information on the cost of dishes was used for the latter approach despite the pitfalls of doing so.²⁵⁰

An interesting attempt at SKA cost modelling was made in 2007 in SKA Memo 90,²⁵¹ in which John Bunton (Australia Telescope National Facility) derived scaling for cost vs frequency, using collected historical information. Although the analysis was thorough, Bunton concluded “The author does not have a great deal of confidence in this result and, to confirm it, significantly more data are needed.”

While cost-modelling of dishes was unreliable, cost models of the entire SKA system were an essential tool for understanding system trade-space, although not with high precision. (For a more general discussion of system trade-space, see Sect. 6.2.2.7 and SKASUP6-4). The radio astronomy community had actual construction experience with other components of the system, such as receivers, correlators, beamformers, operations and software. These other components have a direct influence on the number of dishes, hence their diameter and cost. Apart from the aforementioned Benchmark Scenario cost, a competing cost package, SKAcost,^252,253,254 was set up to include PAF technology and WBSPFs,²⁵⁵ but not aperture arrays. Neither tool was complete, and so they were later combined into one tool that could be used for either vision of the SKA.²⁵⁶ It is unclear whether these cost tools had much influence in the final version of SKA1-Mid, but it probably had little influence on dish size, design, or cost. It would be three years before a System Concept Design Review²⁵⁷ forced the SKA to consider a more realistic size for the project, leading to a revised two-phase vision,²⁵⁸ SKA Phase 1 and SKA Phase 2, and a more detailed system description²⁵⁹ of an SKA Phase 1 system. The options available in these tools, including the size and number of dishes, had already been nailed down and the tools at that point were no longer needed.

The ‘rubber hit the road’ in 2009. To keep the US in the project, the SKA needed to be represented and highly rated in the upcoming report from the ASTRO2010 Decadal Survey of astronomy²⁶⁰ (Blandford et al., 2011). Among other things, the SKA needed to put forward a costed design, even though the SKA project was not sufficiently mature to undergo such detailed scrutiny.

Despite the risks (see SKASUP6-10²⁶¹), the TDP, supported by the SPDO,²⁶² directly used dish-cost estimates from the ASKAP project to underpin costs, as that was the best information available at the time. The 12-m ASKAP antennas were designed and fabricated by the CETC54 company in China and built on site in Western Australia. Their unit structural cost was expected to be €160 k in 2009 (€1400/m²).^263,264 In hindsight, this basis of estimate was difficult to ascertain for a variety of reasons. This led to scepticism, especially in the US, that the total cost estimate of $1767 million for the full SKA was reliable (see discussion in Sect. 4.5.3), leading to the eventual withdrawal of the US from the project.

By mid-2010 the 15-m DVA-1 design had momentum (see Sect. 6.4.5.1). Funded by a combination of TDP, NRC (Canada) and SPDO funds, the construction of a prototype dish began. The reflectors were provided by NRC and a contract was let for the mount and drive system to Minex Engineering Corporation, a small company led by Matt Fleming, who was very active in the SKA and the TDP.

A detailed budget²⁶⁵ was assembled for all aspects of the job: design, fabrication, feeds and low-noise amplifiers (LNA), receiver backend, outfitting and single dish tests (no management overhead). It came to a grand total of US$4.2 million. Of this, the fabrication cost was US$2.26 M, of which US$1.04 M was tooling and setup, leaving US$1.2 M for the dish structure, itself. This was 5.5 times the cost submitted to ASTRO2010, €160 k (US$212 k in 2010). Although a reduction in the DVA-1 cost in production quantities²⁶⁶ would have been substantial, there was no possibility of matching the ASKAP dish costs. A further reduction might have been found by fabrication in a country with low labour costs, but it still would have been a stretch.

So why the huge increase? The conclusion is that it is naïve to give too much credence to cost estimates until there is access to every detail of the design and all the assumptions. Even less credence should be given to scaling. The conundrum is that to obtain accurate costs, irrevocable design decisions must be made, cutting off options.

Cost estimates based on a naïve understanding of the motivations of suppliers leads to poor cost projections, knowledge that is almost impossible to obtain in most supplier situations. An advantage of SPDO paying for part of the work was to have a say in the detailed design and to have access to a detailed cost breakdown, including labour and tooling. This is rarely possible in industry, where cost breakdowns are considered private intellectual property.

6.4.7 Innovations in Sampling the Focal Plane and the Struggle for Phased Array Feeds

The basic theory of Phased Array Feeds (PAFs) has been known for a long time and was described by Rick Fisher²⁶⁷ in 1993, inspired by discussions at the URSI General Assembly in Kyoto, Japan that year. Describing the region around the geometrical focus as a "Fuzzy Focus",²⁶⁸ he pointed out that by sampling the focal region of a dish, all the information needed to construct multiple beams would be available. At the time, however, the necessary signal processing power did not exist.

Sampling the focal plane fields can be done by two basic methods: Method 1) building an array or grid of feed horns, each of which feeds the dish in the way described in the Box in SKASUP6-7²⁶⁹ or Method 2) sampling the electromagnetic fields directly using an array of small antennas fundamentally similar to an aperture array (see Sect. 6.5.1).

Figure 6.19 shows a PAF mounted in the focal plane of a parabolic reflector (dish). It illustrates a scheme that collects information not only from the axial (central) ray path, but also from all ray-paths displaced from the axial ray out to a radius that depends on the details of the design. Thus, the electromagnetic field pattern in the focal region (i.e., the Fuzzy Focus) emanates from a large angular region on the sky around the axial direction. The PAF consists of an array of small antennas (PAF elements, each about half a wavelength in size) that senses the off-axis fields. Their outputs can be combined to provide an image of a patch of sky several beams wide, rather than just one beam area. A more detailed explanation is provided in the Box in SKASUP6-11.²⁷⁰

Fig. 6.19

An illustration of the principles of PAFs

PAF-equipped dishes also have the advantage that the field-of-view is approximately constant over its useful operating frequency range. This is in contrast to dishes with single-pixel feeds, whose areal field-of-view is inversely proportional to the square of frequency.

Although PAFs are an attractive concept, there are several issues that have made its widespread implementation in arrays difficult:

• Each of the antenna elements in the PAF requires an LNA. In a standard radio astronomy configuration, the LNAs are cryogenically cooled to reduce amplifier noise.²⁷¹ But for a PAF at the frequency of the hydrogen line (1420 MHz) whose diameter is more than one metre, this would require an impractically large, heavy cryostat. Hence, since the LNAs on large PAFs must operate at ambient temperature, the amplifier noise can be ten times that for cryo-cooled LNAs. In a comparison of PAF Survey Speed sensitivity²⁷² (see Sect. 6.4.1 and SKASUP6-7²⁷³) with the traditional system, a dish with a low-noise LNA, scanning the same area as the PAF sky-coverage, can image with higher sensitivity than a PAF in the same observing time.

• In any parabolic dish, the maximum focussed power is available on the principal axis. Off-axis power is lower and more spread out. This means that efficiency is lower for beams formed from PAF antenna elements in off-axis positions.²⁷⁴ Moreover, the off-axis beam shapes vary slightly from one another, complicating calibration and beamforming. Efficiency loss lowers the signal-to-noise ratio of data from those beams. And finally, there are strong coupling effects between the PAF antenna elements, which must be considered in forming the beams. This means that more inputs from the elements must be summed than if the antenna elements were receiving independent information.

• Another area of concern is the cost of signal and data processing (see Fig. 6.19). This system component is not required in a traditional design where single ‘analogue’ feed horn does the same job for one beam.

In the long term, it may be possible to overcome or ameliorate these issues. For example, the cost of digital signal processing (DSP) continues to fall, and such processes now consume less power. Also, uncooled LNAs have been developed that are much less noisy than those from a few years ago, but LNA progress is much slower than observed for DSP. Research workshops continue to be held (see Sect. 6.4.7.4) to utilise these developments with the goal of making PAFs more competitive.

PAFs are easiest to implement on large single dishes where there is the space and weight-bearing capacity at the focus to mount cryogenically cooled PAFs (cryoPAFs) (e.g., (Jeffs et al., 2008; Cortés Medellín et al., 2015; Roshi et al., 2017)). This investment is much more worthwhile on these telescopes than on arrays of smaller antennas.

However, there is a corner of radio astronomy discovery space inhabited by uncooled PAF-equipped arrays of radio telescopes like ASKAP where there is more of a premium placed on sky coverage than sensitivity. With the large sky coverage, ASKAP and other telescopes in this category (e.g., CHIME (Amiri et al., 2018)) have discovered extraordinary tracers of far-away astrophysical events—powerful radio transients of extremely short duration—now known as Fast Radio Bursts (FRBs). Because they are strong and originate randomly in the sky, instantaneous sky coverage is more important than sensitivity (see Sect. 6.2.2.8).

6.4.7.1 PAF Developments in Individual Institutions

Over many years starting in 1998, significant research and prototyping effort was devoted to PAFs in Canada, Australia, the Netherlands and the USA which we briefly summarise in the following sections. A brief history of these developments from the Australian perspective has been written by Ron Ekers and John O’Sullivan that describes the transition from the successful development of multi-beam receivers on the Parkes radio telescope (method 1 in the previous section) to discussions at CSIRO in the late 1990s of PAF designs (method 2).²⁷⁵ Some of this work continues at the time of writing. Each institution had different motivations, but all were intent on applying them to the SKA telescopes in some way.

6.4.7.1.1 Dominion Radio Astronomy Observatory (DRAO, Canada)

DRAO/National Research Council began planning the Large Adaptive Reflector (LAR)²⁷⁶ in about 1996, which required a large PAF suspended by an aerostat to feed a huge prime focus reflector on the ground. This scheme could not operate efficiently without the ‘programmable beam-shape’ capability inherent in the PAF concept. Hence LAR success was completely dependent on the success of PAFs.

When the SKA Reference Design (see Sect. 6.2.1.3), including the choice of small diameter dishes was adopted in 2006, the LAR project was wound down, but interest in PAFs continued as a potential contribution to the SKA. The initial investigations had already begun to reveal some of the limitations of PAFs for radio astronomy (Veidt & Dewdney, 2005b). By that time, other groups were active in PAF research and DRAO’s inclination was to collaborate but continue their own program. DRAO collaborated actively with NRAO and the Université Catholique de Louvain in Belgium to understand whether PAFs could play a role in upgrading the Very Large Array (VLA) in New Mexico.²⁷⁷

But it was obvious that such complicated radio frequency (RF) systems could not be understood by paper analysis and theory alone. In 2005, DRAO obtained funding to build a Phased Array Demonstrator,²⁷⁸ known as PHAD²⁷⁹ (Veidt & Dewdney, 2005a), collaborating with the University of Calgary.

The PHAD program was created to develop a fundamental understanding of the capabilities and limitations of phased-array feeds on reflector antennas, and to answer some key questions about the design of PAFs^280,281 (Veidt et al., 2011). It was a modest-sized, engineering demonstrator consisting of a 200-element array of Vivaldi antenna elements.²⁸² The system was designed for flexibility and quick turn-around of results.²⁸³ In a related program, the University of Calgary pursued an even more advanced version of focal plane sampling using 3D space-time plane-wave filtering techniques.^284,285 For further details see SKASUP6-12.²⁸⁶

DRAO continued an active role in collaborations with the other institutions until 2015, when the Herzberg Institute proposed to build a large cryogenically cooled PAF,²⁸⁷ although this never came to fruition.

6.4.7.1.2 CSIRO (Australia): Development of ASKAP

The CSIRO group in Australia recognised PAFs quite early as an emerging technology for radio astronomy, a compromise design approach between passive collecting area exhibited by reflectors and active collecting area exhibited by aperture arrays. PAFs could combine well understood technology for both. In contrast to the conservative investigations in Canada, the well-funded CSIRO group eventually ‘bet the farm’ on PAFs as the best approach to achieving the sensitivity (survey speed) goals of the SKA. In 2002 CSIRO and ASTRON collaborated on an experiment to use an aperture array tile as a feed for a Luneburg Lens.²⁸⁸ In 2004/2005 PAFs became the underlying technology driver for the series of prototype array proposals. Prototyping began in 2005 with a series of test interferometers, the New Technology Demonstrator (NTD) (Hayman et al., 2008), followed by the xNTD.²⁸⁹ But these were overtaken by events and not actually completed.

Not surprisingly, the test sites in the Sydney area were badly contaminated with RFI, which led to a decision to carry out the testing on the remote site proposed for the whole SKA in Western Australia. This also satisfied a quasi-political need to show that CSIRO could effectively use the site and led to the test project becoming an entity encompassing almost all of Australia’s SKA ambitions. Progressing through a variety of names,^290,291 by 2007 it had grown into a 36-antenna telescope proposal, now renamed the Australian SKA Pathfinder (ASKAP), and received $A95M in construction funding (see Sect. 4.3.3.1). At that point it had all the accoutrements of a formal project.²⁹²

Although the TDP in the US was not very interested in PAFs, they were charged with tracking the progress of dish technology, and ASKAP was reported upon and promoted in Antenna Working Group (AWG) meetings²⁹³ on both technical and scientific fronts.

By this time there was a great deal of confidence at CSIRO in PAFs and how to build their supporting infrastructure, such as beamformers, for a reasonable cost. For example, with considerable foresight in 2004, it was clear that many thousands of RF signal chains would be required for the SKA, and so CSIRO embarked on plans for a Wideband Complementary Metal-Oxide-Semiconductor (CMOS) Integrated Receiver (Jackson, 2004), an integrated circuit that could greatly reduce the cost of these components.

While most competitors were focussed on Vivaldi antenna elements, the CSIRO group developed an array of patch antennas for the ASKAP PAFs, which they reported on in 2006^294,295 (O’Sullivan et al., 2008). These look like a chequer board printed circuit (Fig. 6.20), connected to amplifiers located below the circuit board through perpendicular transmission lines at the corners of each patch. The development of sky-mount antennas (see Sect. 6.4.4.1) solved a major issue with PAFs, fixing the orientation of the array of beams on the sky while the Earth rotates. By 2009 the ASKAP proposal was transformed into a major astronomical facility (DeBoer, 2009).

Fig. 6.20

An early version of the ASKAP chequerboard array being tested in an anechoic chamber by John O’Sullivan, the designer. Credit: CSIRO Radio Astronomy Image Archive CRAIA-SKA009

A collaboration on ASKAP was initiated in 2006 between CSIRO and the Canadian institutes involved in this aspect of SKA design.²⁹⁶ Canadian scientists were very interested in a Southern Hemisphere telescope at deci-metre wavelengths that promised observing in such a short timeframe. A science proposal with a Canadian flavour²⁹⁷ was developed and discussed widely, and significant instrumental variations to improve ASKAP’s capabilities were included.²⁹⁸ However, at the same time, concerns were building²⁹⁹ that such a major effort would deflect from the larger goal of SKA participation. Not being able to negotiate a solid contribution to ASKAP, the National Research Council in Canada eventually abandoned the collaboration in favour of putting as much effort as possible into the SKA itself.

An additional collaboration between CSIRO and SKA South Africa, called CONRAD (CONvergent Radio Astronomy Demonstrator³⁰⁰) was established in 2007 to address SKA computing challenges, particularly for the software required for very wide-field imaging. This turned out to be a fruitful exercise that continued for several years.

Adoption of PAFs for the SKA hit a major bump in the road around 2010; there was a building consensus in a series of meetings organised by the TDP that the optical design for the SKA dishes had to be an offset Gregorian (dual reflector) antenna.³⁰¹ Moreover, in the quest to maximise sensitivity, the SKA dishes were to be ‘shaped’ (see Sect. 6.4.5.1). This greatly increases the gain of the antennas at the expense of lower gain when the feed is displaced from the central ray (off-axis performance). This obviously affects the sensitivity of the off-axis elements of a PAF array. The energy is still present in the off-axis focal region, but it is more spread out and requires a larger PAF to capture it efficiently.

In comparison, the ASKAP dishes are relatively simple prime-focus, unshaped designs,³⁰² ideal for PAFs because off-axis performance in this design falls off relatively slowly. In the case of a PAF feed, shaping is not necessary because in principle the same effect can be created by the way that the sampled data from all the PAF elements is combined to synthesise a dish beam.³⁰³

In 2008, anticipating the trend towards offset Gregorian designs, the CSIRO began a series of investigations as to how well PAFs would work on these reflectors. This work continued for several years, primarily to preserve the option to construct a third SKA telescope in Australia—SKA Survey (see Sect. 6.4.7.5). SKA Survey was proposed in the 2013 Baseline Design³⁰⁴ (see Sect. 6.4.7.5) as a much larger follow-on to ASKAP. However, designing specific dishes for SKA Survey would have been quite expensive; the cost would be much lower to simply adopt the SKA-Mid design and to equip them with PAFs for SKA Survey. This led in two directions, one to influence the SKA design to be as PAF-friendly as possible, and the other to address the performance and cost issues arising from equipping SKA dishes with PAFs. A PAF-friendly design would be unshaped, but the other SKA partners were not willing to accept the concomitant lower sensitivity.

A presentation at the Dish Conceptual Design Review in 2011 discussed the possibilities and concerns to be addressed.³⁰⁵ One of the greatest concerns was field rotation for an offset Gregorian dish design; the option of a third axis was not possible because the dish is not rotationally symmetric.³⁰⁶ In 2012 the design of the two telescopes (dubbed ‘SKA SPF’ and ‘SKA Survey’) were broadly compared.³⁰⁷

Further developments on PAFs for the SKA are summarised in SKASUP6-13.³⁰⁸

6.4.7.1.3 ASTRON (The Netherlands)

Although the SKA-related focus in The Netherlands was mostly on dense Aperture Array (AA) technology (see Sect. 6.5.5), there was a significant interest and effort devoted to PAFs, beginning in 2002 when ASTRON sent an AA tile to CSIRO for evaluation as a PAF.

Driving the development of PAFs in The Netherlands was extending the life of the Westerbork Synthesis Radio Telescope (WSRT). The WSRT had been a workhorse of radio astronomy for decades, but the development of newer radio astronomy initiatives in The Netherlands indicated that a change was needed. The combination of general expertise in phased array technology and a scientific interest in deeper wide-field observations of atomic hydrogen gas (HI) pointed almost directly at PAFs for the WSRT. Moreover, ASTRON experts in electromagnetics had been closely involved with PAF theory (e.g., (Ivashina et al., 2011)). They collaborated extensively with colleagues at CSIRO, Brigham-Young University (BYU) and the National Radio Astronomy Observatory (NRAO) in the USA, and DRAO in Canada.

Serious development began with a PAF prototype, DIGESTIF, a small array of Vivaldi antenna elements, mounted at the focus of one of the WSRT reflectors. Interferometric imaging was demonstrated in 2009³⁰⁹ by using other antennas in the WSRT array to correlate with a beam formed from a PAF-equipped antenna. APERTIF (APERture Tile In Focus), the full system, deployed on 12 of the WSRT antennas, opened in September, 2018. An updated technical description is contained in (van Cappellen et al., 2022). A surprising and important technical result was a reduction in so-called standing waves,^310,311 multiple reflections of the radio waves between the focus box and the vertex of the reflector.³¹² This effect had long been a bugbear for spectral line observers.

Although the uncooled APERTIF PAFs were a good fit for their purpose on the WSRT, they were not suitable for modern antenna arrays like those being built for the SKA, mainly because of system noise.

From the perspective of the SKA, the influence of PAF development in The Netherlands came mainly through research into electromagnetic fundamentals rather than direct involvement in their implementation on SKA antennas.

6.4.7.1.4 US National Radio Astronomy Observatory (NRAO) and Brigham-Young University (BYU)

Stemming from long-standing interest in PAFs,³¹³ NRAO and BYU carried out an energetic academic research program in phased array feeds. Funding for this work was mainly motivated by the ultra-challenging radio astronomy application, but phased array feeds have many other applications in remote sensing and radar as well. BYU provided early theoretical underpinning (e.g., (Warnick & Jeffs, 2008; Warnick et al., 2018)), especially the signal processing for PAFs. They also contributed practical experimental verification in collaboration with NRAO (Jeffs et al., 2008), building a PAF mounted on a 20-m antenna at Green Bank, West Virginia. They also carried out experiments to illustrate the capability to ‘null out’ incoming radio frequency interference (RFI) (e.g., (Nagel et al., 2007)).

These groups understood the need to reduce the noise from PAF receivers and were early developers of cryogenically cooled PAFs (e.g., (Warnick et al., 2011)).³¹⁴ While impractically large and expensive for arrays like the SKA, they are ideally suited for single large antennas such as Arecibo (Cortés Medellín et al., 2015), the Green Bank Telescope and the Parkes 64-m telescope.

6.4.7.2 US Technology Development Program (TDP)

Although the TDP had a mandate in 2007 to coordinate dish development through the Antenna Working Group (AWG) (see Sect. 6.4.5), it paid little attention to PAF developments. The TDP was aware of PAF developments through presentations at AWG meetings and major SKA meetings (e.g., a presentation in March 2008 at the AWG meeting in San Francisco³¹⁵) as well as publications (DeBoer, 2009). This position was partly because of scepticism of PAFs as technically ready for the SKA, and partly because of the influence of the group from the Allen Telescope Array (ATA), which favoured cryo-cooled, wide-band single-pixel feeds (WBSPF). By 2009, this thinking was reinforced by the SPDO, which was anxious to pin down a practical SKA antenna design as soon as possible.

6.4.7.3 Discussions on the Technical Readiness of PAFs, 2006–2015

The technical readiness levels of the wide-field technologies, PAFs (and Aperture Arrays), were the subject of almost continuous discussion in SKA engineering and steering committee meetings for a decade from 2005 (e.g., SKA meeting in Manchester in 2007^316,317).

At a major SKA meeting in Paris in 2006, Peter Hall, the Project Engineer at the ISPO, summarised³¹⁸ the state of development: the newly agreed Reference Design, non-technical issues, the realities of a looming time scale and most importantly, “hard questions” facing the SKA. All together it was a sweeping summary of the state of the technology and the diverse interests of the 12 or so major players. Of course, this especially included the wide-field technologies, AAs and PAFs.

The decision trees contained in this presentation indicated the complexity of design choices facing the SKA; several trees running in parallel were displayed, indicating a major organisational challenge to satisfy the interests of the proponents while also making optimum technical choices. Remnants of this challenge are still present, even as the SKA is well into the build phase.

As described in Sect. 6.2.1.4, a ‘Tiger Team’ was established by the International SKA Steering Committee (ISSC) in March 2007 to revise the SKA specifications and propose a baseline implementation of the SKA, leading to SKA Memo 100. Although the specifications were quite clear, many technical options remained open in SKA Memo 100. For the mid-frequency range (500 MHz to 10 GHz), the options were dishes with single-pixel feeds (wide-band feeds), dishes with PAFs and Aperture Arrays. Sub-groups were assigned to assess the strengths and weaknesses of each option including technical challenges and cost drivers, trade-offs and science implications.

For PAFs, SKA Memo 100 noted that further work was needed³¹⁹ to optimise “complex tradeoffs between many parameters including: focal length-to-dish-diameter ratio (F/D), prime focus or dual reflector, dish diameter, distance of focal plane from the focus, focal plane array size, number of elements, element spacing, upper and lower frequencies and number of beams and beam former inputs.” It was also clear that more experience with operational systems for PAFs was needed. The state of SKA technology development at that time was summarised in a compendium of white papers on potential SKA technology in September 2009,³²⁰ providing a factual background for the Tiger Team discussions leading up to SKA Memo 100. Some months after publication of the Memo, in a compendium of comments made by members of the Tiger Team at a meeting in March 2008,³²¹ it was clear that proponents of AAs were defending their corner with the insistence that AAs would operate very well up to 1 GHz. PAF proponents were less subtle but clearly thought that PAFs were likely to win out. The SPDO emphasised that early choices between PAFs and AAs might be forced by time scale considerations, not allowing much time for deep development.

In the SKA Project Execution Plan) (PEP) of 2011³²² (see Chap. 4 and Sect. 6.2.2.14), PAFs were not part of the Baseline Technology but were included in the AIP along with dense AAs at mid-frequencies and Wide-band Single Pixel Feeds (WBSPF). In this vein “The PAF design and development work will be guided significantly from the results of the PAFSKA program, which encompasses the pathfinder systems APERTIF, ASKAP and PHAD. The aim is to design, fabricate and test a prototype feed system in preparation for procurement for SKA Phase 1.”. This mirrored the approach outlined in November 2010 (SKA Phase 1: Preliminary System Description SKA Memo 130³²³), where there is more detail on the expectations required of the AIP for each of the three technologies.

From those early days and until about 2015, PAF sensitivity (survey speed) in practice was a concern at the SPDO. A theoretical approach in 2010, but considering practical realities, was contained in a conference paper outlining the achievable field-of-view of antennas equipped with PAFs (Bunton & Hay, 2010). In 2012 the first key measurements of early ASKAP PAF sensitivity were available³²⁴ as measured on a 12-m antenna at Parkes. Referring also to the previous paper on the achievable field-of-view, the average survey speed performance of these telescopes, when off-axis aperture efficiency was included, was of great interest at the SPDO.³²⁵ The reduction of sensitivity and additional complexity effects were likely to be exacerbated in SKA dish designs.

In early 2015 a note³²⁶ was circulated disputing the survey-speed projections contained in the SKA Baseline Design and found that the survey speeds of SKA1 Survey and SKA1-Mid were similar.

6.4.7.3.1 Cost of PAFs

Early attempts to develop cost models for PAFs were made in 2006.^327,328 Concluding that the cost of PAFs could be almost as great as the cost of a 10-m antenna structure, the "Cost of PAFs" document³³³ noted (edited for clarity): “Threshold of pain: If the cost per frequency channel were $1000 and the cost of a beam-former were $20, then the cost of a PAF would have been $95,000, about the cost of an antenna. We must do much better than this!”. In hindsight, both PAF and antenna costs were underpriced.

At the Dish Array CoDR in July 2011, now approaching the PEP phase of the SKA,³²⁹ review documents from CSIRO on behalf of the PAFSKA group (see below) contained a cost estimate for PAFs in production³³⁰ of between €200 k and €250 k. This was substantially more accurate, but because SKA1-Survey was never built, it is impossible to know for sure.

6.4.7.4 PAFSKA

PAFSKA was a formal collaboration established in July 2010³³¹ to obtain a consensus agreement on the SKA PAF design within 1–2 years. It kicked off at a PAF workshop hosted by BYU in Provo, Utah in May 2010. Partners were primarily CSIRO, ASTRON, DRAO and BYU/NRAO, with CSIRO as the PrepSKA Lead Institute and coordinator.

A work plan³³² was developed and circulated in July 2010, which contained several milestones from November 2010, ending with the delivery of a production-ready PAF in December 2012. At the Dish Array CoDR in July 2011, PAFSKA presented a series of documents including a concept description for the SKA PAF Sub-system.³³³

PAFSKA was instrumental in holding together a global collaboration of partners with varying institutional goals in the SKA. The individual investigators had a strong interest in overcoming the challenges of this enticing technology. But, despite several workshops, fundamental questions remained (see the next section).

6.4.7.5 Postscript: The SKA1 Survey Telescope

Following the inclusion of an SKA1-Survey Telescope equipped with PAFs in the 2012 SKA site decision, along with a dish array (SKA1-Mid) in Southern Africa and a low-frequency dipole array in Australia (SKA1-Low) (see Sect. 8.6.3), it was included in the official Baseline Design of 2013³³⁴ as a complete telescope sited in Australia. It was foreseen to be a mixed array of 36 12-m diameter dishes from the ASKAP array and 60 15-m SKA Phase 1 dishes, all equipped with PAFs like those for ASKAP. The PAFs covered the frequency range from 650 to 1670 MHz in a single dual-polarised PAF with a 500-MHz instantaneous bandwidth.

An important development occurred in 2014. A more detailed electromagnetic analysis of PAFs^335,336,337 in the optical configurations of the preliminary concept design (Preliminary CoDR) emerged, targeted to participate in the SKA dish down-select process.^338,339 This document was effectively a proposal for a telescope with three PAFs covering three frequency bands starting at 350 MHz.

Although this was a large body of work, it came late in the game, and despite several PAF workshops, reviews and papers in 2012 and subsequent years, fundamental questions remained even in 2015³⁴⁰ although it was recognised that much progress had been made.³⁴¹ Time was running short for the SKA Organisation to make major decisions, and more work on the practical side would be needed to ensure the viability of SKA1-Survey. Three telescopes were looking expensive, and SKA1-Survey was finally dropped from the SKA Baseline Design in a ‘rebaselining exercise’³⁴² in 2015.

In summary, only in Australia did PAFs remain the focus of continued development and implementation as an entire SKA precursor array (ASKAP), including a continuing scientific program. ASKAP was completed in 2019 and has already contributed new scientific results. Australia continues to lead research and development of PAFs, and there is still hope that a breakthrough (or more likely, a series of incremental improvements) will inspire another generation of PAF-equipped radio telescopes arrays.

PAF workshops continued throughout the pre-construction phase, and PAFs remained in the AIP and, post-2021, are included in the SKA Observatory Development Program.

6.5 Aperture Arrays for the SKA

Aperture Arrays (AAs) have a long history in radio astronomy. The earliest radio telescopes were often arrays of simple antennas such as dipoles or Yagi antennas. The first radio telescope, the famous Jansky antenna (Jansky, 1932), operated in a similar way; the principles have been known for a long time. The AA terminology is more recent and refers to a much more sophisticated use of the same basic idea.

As described in Sects. 3.2 and 3.3, in the period beginning in 1993 with the formation of the Large Telescope Working Group (LTWG) and the later signing of the Memoranda of Agreement in 1996 for collaboration on technology studies³⁴³ and in 2000 to form the International SKA Steering Committee (ISSC), all the countries involved were motivated to explore innovative technologies, particularly for ‘receptors’ or ‘concentrators’, that could enable the realisation of the SKA.

AA technology seems seductively simple, in principle, but is actually sophisticated and complex. The big push in the early part of this century was to develop dense AAs (see Sects. 6.5.5.3 and 3.2.6.1). Dense AAs promised in principle to revolutionise radio astronomy around the all-important wavelength of the neutral hydrogen line, 21 cm (see Sect. 5.10).

AAs were and are being used in other radio science and engineering disciplines such as radar, but these applications are typically much less demanding than those for radio astronomy (van Ardenne et al., 2009).

To understand the long history of AA development in the SKA and why individuals and whole institutions took heartfelt (sometimes entrenched) positions on the design, it is important to understand some of the basic technology. This is described briefly in the next section.

6.5.1 Basic Technology of Aperture Arrays at Low- and Mid-Frequencies

AAs tailored for the SKA have taken the form of large, horizontal arrays consisting of 100 s to many 1000s of simple antennas, referred to as array elements. Each array, known as a station, forms one or more beams, each of which are equivalent to a lighthouse beam working in reverse, receiving radiation rather than transmitting radiation. The outputs of these beams are delivered to a centralised facility where they are correlated. In this sense each station acts like a dish, except that in principle, AAs can produce multiple beams in arbitrary directions. An array of AA stations can be seen as equivalent to re-using the collecting area of the telescope for making multiple observations at once.

In theory, AAs offer great flexibility. Multiple beams can be formed, which in principle can cover most of a hemisphere simultaneously. In addition, the shape of beams can be adjusted to form nulls³⁴⁴ in the directions of strong sources of radio frequency interference. There are no moving parts, in contrast to reflector antennas.

Figure 6.21 is a simplified view of an AA station for the SKA that applies for both low- and mid-frequencies. The depiction in this case is more like that used for SKA-low than that proposed for SKA-Mid but shares most of principles described here. Each of the red and blue antenna elements is represented here as a kind of dipole with triangular arms, somewhat less than a wavelength in size. Each element has two dipoles orthogonal to each other so that they can receive both polarisations of the incoming radio waves. These antennas are sensitive to almost any direction in the sky. Their outputs are amplified (not shown) and delivered to a device that can introduce programmable amounts of delay into each signal. A wavefront is shown in the figure as radio waves emanating from a direction which is chosen by programming the delays to compensate for the geometrical delay between each element and a reference point, normally the centre of the array. An example of the geometrical delay for one antenna is shown in the figure. After the delays have been introduced, the signals are all aligned for the desired direction. When they are arithmetically summed in the next step to form one output signal, the peak occurs for signals in the desired direction. This can be done to form a beam anywhere above the horizon. Beyond this simple concept are a variety of effects that are much more complex and require significant performance trade-offs.

Fig. 6.21

A simplified view of an aperture array station for the SKA. The array elements are red and blue, used to denote the two antennas oriented to receive orthogonal polarisations

The ability to form multiple beams means that the instantaneous field-of-view can be increased just by forming more beams from the same collecting area. Survey speed (see also Sect. 6.4.3.1 and SKASUP6-7³⁴⁵), a fundamental metric of telescope sensitivity, is used to capture the time taken to survey an area of sky to a given sensitivity level. The metric is (A_e/T_sys)² ∙ Ω, where Ω is the area of sky covered by the beams and A_e/T_sys is the sensitivity of the collecting area. Reuse of the collecting area is a powerful concept, and if beamforming could be formed at negligible cost, then AAs could easily achieve very high survey speed. However, beamforming cost is not negligible, and one must evaluate survey speed based on the full system cost.

Dense AAs (see Sect. 6.5.5) contain closely packed elements in a rectangular or hexagonal array so that the elements are approximately a half-wavelength (λ/2) apart (Fig. 6.24). This means that all possible spatial scales are sampled by the array, up to the limit set by the size of the array. AAs whose elements are farther apart than λ/2 are so-called sparse (sparsely sampled).

The sensitivities of sparse and dense AAs differ: In a sparse array, the elements are far enough apart to act independently, and the collecting area of the individual elements is proportional to the square of the wavelength (λ²). Thus, the collecting area of the array is proportional to Nλ², where N is the number of elements in the array. On the other hand, because the elements of dense AAs are strongly coupled, the collecting area is proportional only to the area of the array, regardless of the number of antenna elements. Over a wide frequency range, it is possible that a single array may be dense at long wavelengths but sparse at shorter ones. Setting the transition frequency, which occurs approximately when the elements are a half wavelength apart, involves a complex trade space containing the design of the elements, the frequency coverage desired and the sensitivity.

Another design choice for the SKA is the size of AA stations. For a fixed total collecting area, for example as specified in SKA Memo 100, larger stations will require fewer of them. If the stations are too large, then there may not be enough of them to support accurate, high dynamic range imaging. If they are too small, the system cost will rise. The station beam area is proportional to (λ/D)², where λ is the wavelength and D is the station diameter. If there is a requirement to observe a fixed area on the sky (field-of-view) or to maximise it, then it may require several beams to cover the required area on the sky. Each beam requires a beam-forming apparatus, whose cost is also proportional to the number of antenna elements in the station.

The station size affects cost in other ways as well: each antenna element requires at least a Low Noise Amplifier, which consumes power. Transmission of their output signals also consumes power. In general, the power consumed by a station is proportional to its area.³⁴⁶

Finally, as with all radio telescopes, system noise³⁴⁷ is crucial. Figure 6.22 shows a dramatic natural phenomenon. The sky is exceedingly bright at long wavelengths (low frequencies) but almost completely dark at frequencies greater than about 500 MHz.³⁴⁸ The other typical source of noise is from Low Noise Amplifiers (LNAs). LNAs that operate at ambient temperatures contribute about 40 K of noise, which is insignificant at 50 MHz but much higher than sky noise at 1000 MHz. The sensitivity of telescopes is proportional to (A/T_sys), where A is the total collecting area and T_sys is the system temperature. At low frequencies it pays to maximise collecting area but at high frequencies it pays to minimise LNA noise. This is a key reason why it is difficult to imagine one telescope covering the entire SKA frequency range. It is also the reason why AAs are more suitable for wavelengths longer than about a metre (300 MHz) than for shorter wavelengths, although AAs at higher than 300 MHz are certainly possible.

Fig. 6.22

The average radio brightness of the sky at SKA’s AA frequencies. Note that the vertical scale is logarithmic and the horizontal scale intervals are in factors of 2

SKASUP6-17³⁴⁹ contains a summary of the AA challenges, both fundamental and cost-related. In terms of the SKA specifications as they were known in the 2004–2007 timeframe, the decision space looked like the following:

• Frequency range? (How many AAs?)

• Station size?

• Reconfigurability and flexibility?

• Antenna element?

• Sparse or dense; if sparse, what transition frequency, if any?

• Field-of-View? (How many beams?)

• Cost per square metre of collecting area?

• Cost per square degree on the sky?

• Operating cost?

• Sensitivity and instrument noise?

Optimising this enormous trade space was the task at hand, but the results would not appear for years.

6.5.2 AA-Focussed System Architecture

In the period from before the SKADS program (see Sects. 6.2.1.6 and 6.5.5.2) began in 2005 (see Sects. 3.3.3.4.3 and 6.2.2.1) until the SKA System CoDR in early 2010, when dense AAs were included in the AIP but not in the system baseline (see Sects. 4.6.2 and 6.2.2.9), there was an underlying assumption that there would be an integrated approach to AAs at the system level (i.e., one beamforming ‘engine’ would handle inputs from both sparse AAs at low frequencies as well as dense AAs at high frequencies). It would also provide a central correlator system for AA beams and for dishes. This approach is clearly efficient since the basic beamforming process is the same for both. In principle, it could also provide the flexibility to process more or fewer beams from either of the AA arrays.

Arnold van Ardenne included a version of this architecture (Fig. 6.23) in a presentation of the Aperture Array Verification Program (AAVP)³⁵⁰ to the SKA International Engineering Advisory Committee (IEAC) in April 2009 in which he declared that “AA’s are essential for SKA Science”, meaning that this architecture and structure would meet SKA Memo 100 specifications (see Sect. 6.2.1.4).

Fig. 6.23

AAVP system design. Credit: Andrew Faulkner and the AAVP Program

Because of the shared processing for both dense and sparse AAs, it was difficult to see how to map this architecture into a footprint on the ground—the array configuration. For example, the distribution of both sparse and dense AAs, as well as dishes, were supposed to be highly concentrated at a single centre. It would have been difficult to find room for everything, exemplified by leaving AA-Low arrays out of publicity pictures. This led later to a multi-core approach (see Figs. 6.6 and 6.7), which if this integrated architecture were to be retained, would mean long-distance data transmission networks. Moreover, the subsequent dual-site decision in 2012 completely changed all the underlying architecture assumptions.

6.5.3 Design of SKA Low Frequency Aperture Arrays: Compromise and Convergence

Aperture Arrays go back to the earliest days of radio astronomy. After his much earlier observations with dishes, the pioneer radio astronomer, Grote Reber, built an array in Tasmania at the extraordinarily long wavelength of 144-m (frequency of about 2 MHz) in the 1950s and 1960s.³⁵¹ Other pioneering array-type telescopes for shorter radio wavelengths were built at several observatories (e.g., (Caswell, 1976; Erickson & Kuiper, 1973; Braude et al., 1978; Roger et al., 1999)). Because their antenna elements were tightly coupled, they were dense AAs, although the term was not used then.

Interest in detecting neutral hydrogen at cosmological distances using a telescope with about a square kilometre of collecting area was sparked by Swarup, Braun et al. and Wilkinson in the late-1980s (see Sect. 2.4.1). This idea and science discussions in the late-1990s roughly set the scene for the low-frequency limit for SKA-low, which began at 100 MHz, and changed downwards to 70 MHz in 2005 and is now 50 MHz.^352,353 The high frequency limit, which depended on defining a cross-over point between SKA-Mid and SKA-Low, was initially about 350 MHz but remained fluid for a long time afterwards.

The prospect of discovering highly redshifted hydrogen was one of the ideas that also propelled LOFAR as a potential mega-project at low frequencies (see (van Haarlem et al., 2013) for a sketch of its early history), which emerged as a named project in 2000 (Bregman, 2000). However, for a variety of reasons the original consortium dramatically trifurcated in 2004 into the modern LOFAR telescope in Europe, the international Murchison Widefield Array (MWA) in Australia and the Long Wavelength Array (LWA) in the US (see Sect. 3.2.6.1). The builders of LOFAR and MWA (Lonsdale et al., 2000) have also been major contributors to the design and construction of SKA-Low. They are pathfinder and precursor telescopes, respectively, to the SKA. One of the originating institutions of LOFAR, MIT/Haystack,³⁵⁴ was responsible for much of the initial design work, especially the antenna arrays.

Earnest paring down of the SKA design began in 2006 with the SKA Reference Design (see Sect. 6.2.1.3). In the ‘sausage-making’ discussions³⁵⁵ in the Tiger-Team leading up to the Reference Design, the so-called ‘Epoch of Re-ionisation Array’ (EoR Array) was given short shrift and most of the emphasis was given to dense AAs at higher frequencies. Although there was acknowledgement that an array spanning <100 to 300 MHz was needed, there was very little discussion of the science potential in the upper end of that frequency range. The EoR Array was not even included in the publicity image of the Reference Design. It is ironic that of all the technologies being discussed at the time (2005), only the EoR Array and dishes were included in the final design of SKA1. Nevertheless, these were optimistic, exciting times when huge technical strides in the designs of radio telescopes were thought to be possible.

In the Reference Design document, the EoR Array was not included in SKA Phase 1³⁵⁶ because “10% SKA has the same collecting area as LOFAR at 0.1 GHz, and therefore is not expected to add substantially to EoR knowledge unless the EoR signal is primarily to be found in the Frequency Modulation (FM) bands which LOFAR cannot observe.”³⁵⁷ (Note that at that time, SKA Phase 1 was considered as just a milestone on the way to constructing the full SKA).

In SKA Memo 100 (see Sect. 6.2.1.4), the assumed implementation of AAs included two sparse AAs, one covering from 70 to 250 MHz and the other one up to 450 MHz. These two AAs would have used two ranks of dipole-like antennas to cover the range. There was also an acknowledgement that perhaps Vivaldi antenna elements could cover the entire range with one array. But neither array was eventually chosen for SKA Phase 1.

Performance requirements were extraordinarily ambitious. For example, in 2008 Joe Lazio, the SKA Project Scientist, presented AA-Low specifications to the SSRC,³⁵⁸ which included sensitivity of at least 4000 m²/K for zenith angles up to 45°, polarisation purity of −30 dB (1 part in 1000) over a wide field-of-view, and frequency resolution of 500 Hz up to a frequency of 250 MHz. These were widely accepted at the time and did not ‘raise any eyebrows’, even in the SKA engineering community. The polarisation requirement was especially difficult because dipole-like antennas inherently change polarisation in off-axis directions.

Detection and imaging of neutral hydrogen in the eras of Reionisation and Cosmic Dawn remained the science drivers for SKA-low³⁵⁹ (see Sect. 5.5.19) and formed the basis for the SKA1-Low part of the Baseline Design³⁶⁰ in 2012/13.

In the AAVP program as defined in 2009, INAF and ICRAR³⁶¹ were assigned the development of low-frequency specific AA components (Fig. 6.32), but it was expected most development would entail adapting designs from LOFAR.³⁶² However, other approaches were still being considered throughout the AAVP period, and there was some frustration at the slow pace of convergence on a design. At the AAVP meeting Perth in 2011,³⁶³ Andre Gunst, who at the time was working at the SPDO, frustrated at the pace, pointed out that SKA system requirements and an SKA high-level design were in place already and LOFAR could just be scaled to build AA-Low, without further design iterations.

Nevertheless, the AAVP project engineer, Andy Faulkner showed in a presentation at the AAVP meeting in Medicina³⁶⁴ that many key choices were open,³⁶⁵ even in 2012. But by that time, dense AAs were no longer part of the Baseline Design, and all the options were tailored for SKA1-Low.

As noted in the example above, a characteristic of this period was to avoid making design choices that would constrain the potential flexibility of AAs. However, the pressure of the imminent end of the PrepSKA program and the beginning of pre-construction, meant that choices did have to be made. These choices were embodied in the Baseline Design document of 2013, which set the project on a course to follow, and which remained broadly stable up to SKA construction start in 2021.

6.5.3.1 SKA-Low Array Elements

A great deal of effort was put into studies of the design of antenna elements for SKA-low.³⁶⁶ Although other alternatives were studied, the choices most intensively studied were: (1) Simple dipoles and their derivatives (the default), (2) Vivaldi antennas,³⁶⁷ (3) Log-periodic dipole antennas,³⁶⁸ and (4) Conical antennas. Further detail on these antennas is given in SKASUP6-18.³⁶⁹

Another question discussed early in the pre-construction phase (October 2012), was whether to have two arrays of ‘simple’ dipole arrays or one array of log-periodic or Vivaldi antennas covering the whole SKA1-Low frequency range from 50 to 350 MHz.^370,371 Log-periodic elements were ultimately selected for SKA1-Low in the Baseline Design in early 2013 because they could cover the frequency range from 50 MHz to 350 MHz, obviating a dual array approach.³⁷² At the upper frequency, this choice was made easier by the choice of 15-m dishes for SKA-Mid which could provide acceptable performance down to 350 MHz. However, the choice of log-periodic dipole antennas remained controversial for years (see SKASUP6-18).

6.5.4 SKA-Low Stations and Array Configurations

Because the flexibility of the AA concept provides an enormous number of design options, both the configurations of the antennas within stations and the configurations of the array of stations generated considerable discussion. The issue of flexibility was not fully settled until 2016.³⁷³

In the 2010/11 period, discussions in the AAVP, which was leading the SKA-low design, led to an array design containing 50 very large (180-m diameter) stations containing about 10,000 antenna elements each. This diameter was carried forward to the AA CoDR in April, 2011.³⁷⁴ This approach was partly based on being able to calibrate quickly in the face of variable ionospheric distortions, particularly when travelling ionospheric disturbances (TIDs) occurred (Wijnholds et al., 2011). This diameter was carried forward to the AA CoDR in April 2011 together with equivalent information for the dense AAs. However, it was made clear in the CoDR panel’s report³⁷⁵ that treating dense AAs, which were not in the 2010 version of the baseline design (SKA Memo 130), together with sparse AAs for low frequencies in the same Architecture Design Description (ADD) was not optimum for either one.

The clock was running down to the end of PrepSKA, with the expected delivery of a system design, but the debate over the SKA1-Low station size did not end there. Following considerable discussion, especially regarding flexibility, (see SKASUP6-19³⁷⁶) the 2013 Baseline Design chosen was an array of 911 35-m diameter stations, 75% of which were within 1000 m of the centre. The rest were in spiral arms arranged so that the maximum baseline was 100 km. Each station contained 289 log-periodic dipole (LPDA) elements. Further consolidation of collecting area was done in the three spiral arms (see Sect. 6.2.2.10) by arranging clusters of five such stations along each arm (beyond a radius of 2500 m). This configuration is not as good as separating the stations to form more independent samples of spatial frequencies but entails much less expensive provision of power and communication infrastructure.

The design of the configuration of antenna elements within a station became a major research effort in the AAVP. There are several highly coupled design aspects that require multi-way design trades. The technical issues are complex (see SKASUP6-19). They were intensively explored,³⁷⁷ particularly in the UK (El-Makadema et al., 2014), and tested in small prototype arrays.³⁷⁸ A sophisticated array simulator, OSKAR³⁷⁹ (Oxford SKA Radio Telescope Simulator) was developed for radio telescopes containing AAs and is still used for these investigations.

6.5.5 Pursuit of the Ultimate SKA Telescope Design: Dense Aperture Arrays

While sparse AAs had been used for many years, the real prize was dense AAs, scaled to shorter wavelengths or SKA1-Mid-frequencies, as short as 15 cm. Dense Aperture Arrays (AAs) promised to revolutionise radio astronomy.

Figure 6.24 illustrates the variant pursued in The Netherlands, joined later by all of Europe.³⁸⁰ The incoming radio wave induces currents on the Vivaldi antenna elements, shown as colour concentrated around the slot. Low noise amplifiers (LNAs) are connected to each slot. Their outputs are collected in a layer that can deliver all the signals to each beam former. The beams are formed by adding the electrical signals from the elements in a processing network so that delays are equalised for signals arriving from the desired direction (see Figs. 6.21 and 6.24). As with any aperture, the width of the beam is inversely proportional to the size of the aperture, D, in wavelengths, λ (i.e., λ/D) projected on the beam direction. By adjusting the delays, the beam can be pointed over an entire hemisphere. Because the AA station lies in a horizontal plane, beams directed away from the zenith become progressively wider as they approach the horizon. Therefore, practical observations with AAs are confined to angles less than about 50 degrees from the zenith.

Fig. 6.24

An illustration of a Dense Aperture Array Station using Vivaldi antennas

However, there are several practical challenges associated with dense AAs. This are outlined in Box 2 of SKASUP6-17.³⁸¹

6.5.5.1 Mid-Frequency Aperture Array Development Before SKADS

Real interest in AA technology for large-aperture telescopes began in The Netherlands, led by Arnold van Ardenne in about 1993, shortly after ASTRON began investigating innovative technologies for the next large radio telescope. This was followed in 1995 by a significant grant from the Dutch government to bolster the effort (see Sect. 3.2.6.1). Note that slightly later, in the late 1990s, similar technology was being developed in The Netherlands, Australia, and Canada for use at the focus of reflector antennas (focal plane arrays). This aspect is covered in Sect. 6.4.7.

In 1997, ASTRON began designing a series of prototype aperture arrays for the SKA. A major research effort was made, including ascertaining the properties of many different antenna elements, dielectric losses, low-noise amplifier designs, beamforming and developing test apparatus that would be suitable for measuring arrays in an anechoic chamber. The first effort, the Adaptive Array Demonstrator (AAD), was only an 8-element array.³⁸² Of considerable interest was adaptive beamforming whereby weights used in summing the signals from the individual elements could be adjusted in real time to form beam nulls in directions from which radio interference was arriving (see Sect. 6.5.1). This concept was pursued vigorously throughout a series of prototypes.

This prototype was followed by the One Square Metre Array (OSMA) Fig. 6.25, which was much more elaborate. It contained 64 active antennas surrounded by two rows of passive antennas (144 in total) and was designed for the 1.5–3 GHz frequency range. A photo of OSMA on the front cover and a complete summary appeared in the ASTRON 1998 Annual Report,³⁸³ as well as in the first SKA Newsletter,³⁸⁴ indicating the importance of this work.

Fig. 6.25

OSMA in an anechoic chamber at The Netherlands Foundation for Radio Astronomy (later ASTRON). Credit: ASTRON

OSMA was followed in 1999 by the Thousand Element Array (THEA) (Hampson & bei de Vaate, 2001) (Fig. 6.26). This was designed to operate in the 600–1700 MHz range outdoors and to detect radio sources even in the presence of radio interference. THEA was constructed using Vivaldi antenna-elements arranged in 64-element ‘tiles’, which showed considerable promise for very wide bandwidth. Sixteen of these were built and deployed on the Dwingeloo telescope site. Its analogue beamforming network was elaborate: 32 beams that could be directed to any point in the sky, a scheme for adaptive nulling and full computer control (van Ardenne et al., 2000; Smolders & Kant, 2000; Kant et al., 2000). In 2002, observations with THEA detected Galactic HI³⁸⁵ (Wijnholds et al., 2004) and could track GPS satellites in orbit. This was described in detail, in Dong Xiao's thesis from the Eindhoven University of Technology (TU/e).³⁸⁶

Fig. 6.26

Left: The THEA Tile with the radome removed Credit: ASTRON. Right: An array of 16 THEA tiles, four active tiles surrounded by 12 passive tiles on the Dwingeloo site, flanked by Harvey Butcher, ASTRON Director at the time. Credit: ASTRON

Effort began to develop innovative beamforming techniques using photonic devices (e.g., as proposed in a poster paper by Peter Maat³⁸⁷). This continued into the SKADS era (see next chapter).

In general, an extraordinary effort was devoted to dense AAs at ASTRON and associated universities in The Netherlands, in the period up to the funding of the SKADS program in 2005. At that point work continued at the European level but led in The Netherlands. Apart from the developments described above, a highlight of the activity was work on developing adaptive beamforming for radio telescopes utilising AAs. This is summarised in Dong Xiao's thesis and in some of the publications referenced above.

6.5.5.2 SKADS: Mid-Frequency AAs

Section 6.2.1.6 describes the inception of the SKA Design Study (SKADS)) program, funded by the European Commission’s Sixth Framework Program (FP6).

The technical focus of SKADS was the continued development of mid-frequency AAs for the SKA, but its overall goals went beyond this remit: make the case for AA technology including readiness for production, produce science requirements, develop a full-blown architecture and project plan, estimate a cost and organise industrial participation. Although these goals were only partly achieved, SKADS left a large technical legacy. The technical component was led by ASTRON but continued with significant collaborators from the UK and France, and connected projects in Australia and Canada, where some of the results of this work fed into the development of phased array feeds (PAFs) for dishes (see Sect. 6.4.7).

At mid-frequencies, the AA-inspired design promoted a very wide-ranging vision of a telescope with key properties delivered concurrently: wide field-of-view, high resolution, wide frequency range, wide bandwidth, high sensitivity, dual-polarisation capable, wide time-domain capabilities, and the flexibility to observe multiple targets concurrently at will, all at modest cost. In general information theory terms, the amount of information available is proportional to the product of all these terms except ‘flexibility’. An efficient, error-free telescope would deliver an unprecedented amount of data to be processed and understood, and such a telescope would satisfy almost any science requirements. The question is whether a practical implementation of the AA-inspired telescope could come close to this ideal.

AAs have one advantage that is often overlooked: electronic beamforming can be very rapid. This allows beams to be steered quickly to respond to transients detected on other telescopes or detectors (multi-messenger astronomy). It is also well suited to carry out transient surveys, sampling a variety of astrophysical timescales. This was an emerging aspect of astronomy, the SKA prospects for which are described by Jim Cordes in SKA Memo 97³⁸⁸ and in other places (see Sect. 6.2.2.8).

The program was assembled into a four-year Description of Work (DoW),³⁸⁹ encompassing almost every aspect of radio telescope design and operation. The DoW was organised according to institutions assigned to carry out the work. The University of Manchester was assigned the SKA System Design. The primary AA technology investigations were assigned to ASTRON (Technical Foundations and Enabling Technology) and the University of Cambridge (Networks and Data). The program culminated in the construction of three major prototypes and a beamforming test program.³⁹⁰ Two prototypes, using the AA technology previously developed in the THEA phase, were sited at the Westerbork and the Nançay observatory sites in The Netherlands and France, respectively. The third prototype, 2-Polarisation All Digital (2-PAD), was constructed at Jodrell Bank site in the UK (see Sect. 6.5.5.2.2.).

Of particular interest is Fig. 2 of the DoW, which lists the “Critical Technology Areas”. These illustrate the SKADS overall approach to meeting the challenges outlined in SKASUP6-17.³⁹¹ The topics were:

• 1. Science requirement studies including Configurations, Array calibration, Dynamic range etc.

• 2. Wideband Antenna and Integrated low cost, Front-end technology,

• 3. (Adaptive) (multi) beamforming,

• 4. (Sparse) Array System Design and Engineering,

• 5. High speed processing and Massive Data Handling,

• 6. Array Infrastructure and Network technologies,

• 7. RFI system-level and Mitigation technologies,

• 8. Low-cost Design and Manufacturing,

• 9. Siting and related issues,

• 10. Costing and Specifications, System Design and SKA Plan.

SKASUP6-20³⁹² contains a cross-reference table between the list above from the DoW and the challenges described in Box 2 of SKASUP6-17. While not quite a one-to-one correspondence, this indicates that the proposers were generally aware of the challenges and had outlined an organised approach to investigating them.

By mid-2007, the SKADS had put together a proposal for a system vision for the entire SKA frequency range (100 MHz–20 GHz), which they called the “SKADS Benchmark Scenario”,³⁹³ but which focussed principally on the mid-frequency aperture array, but also included a low-frequency aperture array, modelled on LOFAR or the MWA, and an array of 6-m dishes modelled after the ATA dishes with Wide-band Single Pixel Feeds (WBSPFs) (DeBoer et al., 2004). More detail on this scenario was provided in October 2008,³⁹⁴ following the publication of SKA Memo 100³⁹⁵ (see Sect. 6.2.1.4), which provided a broad consensus on the specifications for the SKA.

The underlying architecture of the Benchmark Scenario (see Sect. 6.5.2) was generally followed until the end of the SKADS program, except for changes in the frequency boundaries between technologies deployed (e.g., the boundary between dishes and dense AAs shifted from 1 GHz to 1.4 GHz (see Table 6.4)).

Table 6.4 Proposed SKADS-SKA implementation from the SKADS White Paper

6.5.5.2.1 SKADS Results (2009)

The SKADS program culminated with a major conference in Belgium in November 2009. The detailed summary of achievements was published electronically in a 410-page volume in the Proceedings of Science.³⁹⁶ However, this was not the end of development of AAs for the SKA, which continued in the AIP program even after dense AAs were not selected for SKA Phase 1 (SKA1) in 2010³⁹⁷ (see Sect. 6.2.2.9).

The SKADS remit was to present a system design for the SKA, focussing on AAs as the primary collectors for Mid and Low frequencies. At Low frequencies, this was not controversial as it had a long history in radio astronomy and indeed this is the technology under construction for SKA1-Low. Also, from a noise perspective, natural noise sources (i.e., sky noise) are the dominant component of total system noise, even for uncooled LNAs, at frequencies below about 250 MHz (1.2 m wavelength).

However, within the four-year program, SKADS could not have been expected to discover or invent new technologies but to create a system that takes advantage of existing and projected technologies to realise the ‘obvious’ advantages of AAs for radio telescopes. An overview of this system was presented in summary form at the SKADS meeting noted above in 2009 by Faulkner, et al.³⁹⁸ and later in more complete form in an SKA Memo.³⁹⁹ Both documents are entitled “SKADS White Paper”. This was delivered thoroughly, presenting the SKA array design shown in Table 6.4, supported by an optical network for gathering data from the stations and dishes, large hierarchical digital beamforming⁴⁰⁰ subsystems, and large correlation and post-processing subsystems. Note the emphasis on very large stations (diameter 56 m) (Fig. 6.27). As illustrated in Fig. 6.28, the AAs were plunging into new technical territory on a large scale.

Fig. 6.27

A slide from the presentation by A. J. Faulkner at the 2008 URSI General Assembly (hba.skao.int/SKAHB-337 Design of an Aperture Phased Array System for the Square Kilometre Array, Faulkner, A. J., et al., presentation at the URSI General Assembly, Chicago, USA, 2008–08) showing the size and extent of an AA station for the SKA. Credit: Andrew Faulkner and the AAVP Program

Fig. 6.28

A grand view of an SKA telescope comprising AA stations. The lower left panel is a view of the arrays of antenna elements in each station. In this rendition each station is about 60 m in diameter. Credit: Swinburne University of Technology, Melbourne, Australia

To examine the results of SKADS in detail is well beyond the scope of this volume. SKASUP6-21⁴⁰¹ contains a compressed approach to explaining how successful SKADS was in meeting the challenges outlined in SKASUP6-17.⁴⁰²

The notes in SKASUP6-21 indicate that although SKADS carried out a thorough-going investigation of design alternatives, many of the technical challenges inherent in dense AAs remained below the level of maturity needed to build a telescope. The SKADS program was certainly one of the highlights of the development of the SKA project up to 2009. Although many of the technical challenges remained in 2009, the huge level of effort provided its own momentum. It was perfectly clear to funding agencies around the world, not just in Europe, that scientists and engineers in large numbers were determined to build a next generation radio telescope and were willing to devote major parts of their careers to do the hard work needed to investigate designs using new technologies. Figure 6.29 is a group picture that illustrates the breadth of interest and the impressive size of the pool of talent working on AA technology.

Fig. 6.29

Participants at the final SKADS conference in 2009 at Chateau de Limelette, Belgium. Credit: S. A. Torchinsky

6.5.5.2.2 Prototype AAs: EMBRACE, 2-PAD and BEST

Although a detailed description of these prototypes is beyond the scope of this volume, a brief outline is included in SKASUP6-22.⁴⁰³ These prototypes⁴⁰⁴ demonstrated that paper investigations could be realised in real hardware. The EMBRACE prototypes⁴⁰⁵ (Torchinsky, 2016) (Fig. 6.30) were capable of astronomical observations (Benthem & Kant, 2012). The 2-PAD⁴⁰⁶ (Fig. 6.31) and BEST⁴⁰⁷ prototypes provided valuable data on the overall performance of SKADS technology and system designs.

Fig. 6.30

Left: The EMBRACE station under a large, curved radome. Right: Inside the radome. Shown are the aluminium Vivaldi radiators in a dual polarisation configuration of about a quarter of the array. Credit: ASTRON

Fig. 6.31

2-PAD installed at Jodrell Bank Observatory. Credit: Peter Wilkinson

6.5.5.3 Dense AAs in the Aperture Array Verification Program (AAVP)

Section 6.2.1.6 describes the transition from the SKADS program to the Aperture Array Verification Program (AAVP), in which work on Dense AAs continued under a new program.

In most respects, the AAVP program continued the structure of the SKADS work packages: AA system design studies (AA-SDS), AA technology development (AA-Tech), build and test AAs on the sky with EMBRACE (A3IV), build and test an all-digital AA (DAAVS), low-frequency component development (AA-lo), and power/infrastructure studies (AA-SEM). Figure 6.32 from van Ardenne’s 2009 IEAC presentation⁴⁰⁸ illustrates the way that assignments and allocations were given to participating organisations. In a prelude to the Aperture Array Verification Program (AAVP) in 2008, Faulkner presented a similar outline.

Fig. 6.32

The AAVP institutional organisation chart as developed in October 2008. Credit: van Ardenne and the AAVP Program

Because frequencies up to 1.5 GHz (or somewhat higher) were to be covered using dense AAs, there was little need in this architecture for large dishes to cover these ‘mid-band’ frequencies. Initially, ATA-style 6.5 m dishes were proposed. Later, somewhat larger ‘simple’ axi-symmetric dishes were proposed.⁴⁰⁹ Later still, 15-m dishes were incorporated, but it became clear that this was not consistent with the AAVP architecture.

As importantly, the integrated nature of architecture and organisational structure was brittle. Everything depended on the assumption that the underpinning technology was successful. When the technology and cost prospects turned out to be more involved than expected, the higher-level structure came under scrutiny. After the SKA System CoDR in February 2010 and the subsequent re-definition of SKA Phase 1, dense AAs were no longer part of the SKA Phase 1 technical concept, and the integrated approach no longer looked appropriate. This did not prevent two large AAVP workshops from being held in late 2010 at Cambridge (UK)⁴¹⁰ and late-2011 at Dwingeloo.⁴¹¹

6.5.5.4 Postscript

Following the System CoDR held in February 2010 and the subsequent decision at the SSEC meeting in March 2010 to down-select technologies for SKA Phase 1 (see Sect. 4.5.2), dense AAs were not included in the baseline SKA Phase 1 technical concept.⁴¹² Recognising the large investment in AA innovations, they became part of the SKA Advanced Instrumentation Program (AIP), along with PAFs and Wide-band Single Pixel Feeds.⁴¹³

The formal AA Conceptual Design Review (CoDR) and a delta-CoDR on the mid-frequency aspects were held in April and November 2011, respectively. The review reports^414,415 continued to outline issues with technical maturity of the mid-frequency (dense) AAs.

Despite a very large effort and significant innovation, the practical implementation of AAs at short wavelengths for radio astronomy has yet to be achieved. SKASUP6-23⁴¹⁶ speculates on whether AAs might have become ‘mainstream’ if the equivalent effort were to have begun at the time of writing.

6.6 Critical Supporting Technologies

Although antenna development has been emphasised in much of the foregoing, aperture synthesis telescopes require much more apparatus to function, the most important of which are discussed in this section.

Arguably the most important devices are the Low Noise Amplifiers (LNAs), which are connected directly to telescope antennas. LNA development had in some cases reached close to the quantum limit (Bryerton et al., 2013), below which no further improvement in noise is possible. Much work was also carried out to produce efficient feeds for dish antennas (see Sect. 6.6.1).

Before 2008 especially, the SKA telescopes were described as Information and Communications Technology (ICT)⁴¹⁷ devices, to emphasise the use of digital technologies in their design. Compared with early radio telescope technologies, which relied more on analogue devices, this represented a great potential improvement in both the performance and cost of the SKA. Peter Hall, then SKA Project Engineer, used the term “ICT engine”, “viewed as a large data transport and processing system” to describe the SKA in general.⁴¹⁸ The emphasis on ICT aspects of the SKA project also attracted talented people from the field, such as John Bunton who was working initially for the CSIRO ICT Centre in Australia.

For broad appeal, the catch phrase “sensor network” was used to describe the LOFAR telescope as a “generic Wide Area Sensor Network for astronomy, geophysics and precision agriculture”.⁴¹⁹ This term was picked up in other documents as well, particularly regarding industrial liaisons.⁴²⁰

These improvements permitted the analogue ‘signal chains’, which start at the antenna and end with digitisation of the signals, to be very short. This confers the advantage of stability and predictability of the all-digital parts of the telescope. Moreover, rapid advances in these technologies provided a route to larger arrays of telescopes than previously thought possible (see points {G} {H} in Chap. 6 introduction). By 2011, cloud computing had also entered the ICT and SKA lexicons.⁴²¹

All these technologies are impacted by RFI and are key components of mitigation strategies in the design of radio telescopes (see Sect. 6.2.2.12).

6.6.1 Feeds, Low Noise Amplifiers (LNAs) and Cryogenics

Because astrophysical radio signals are so weak, designs of radio telescopes go to great lengths to maximise sensitivity. For a given size of telescope (i.e., collecting area), sensitivity is determined by sources of noise, the lower the better (see Sect. 6.4.1). For many years, the most cost-effective approach to improving radio telescope performance was to improve the noise contributed by the LNA, the first amplifier on the antenna, exemplified in a summary from the pre-SKA era by Marian Pospieszalski (Pospieszalski, 1990). These were always cryo-cooled. This strategy was used by the National Radio Astronomy Observatory (NRAO) for many years to maintain its leading role in radio astronomy.

In the early PrepSKA period, it was assumed that cryo-cooling of LNAs on dishes would be too expensive, especially to maintain on the very large number of dishes proposed for the SKA. (It was realised much later that this assumption was too simplistic (see Sect. 6.4.3.3)). Also, during early PrepSPA, PAFs were a hot topic in Australia, The Netherlands and Canada, (see Sect. 6.4.7.1) and PAFs were too large to be contained in a cryostat. Therefore, there was a strong push, which continues at the time of writing, to reduce the noise of LNAs without cryo-cooling. There was also interest in cooling to an intermediate temperature, 80 K instead of 20 K. LNAs for a “Next Generation Very Large Microwave Array”⁴²² was already being considered by Sandy Weinreb at the US National Radio Astronomy Observatory (NRAO) in 1998, and continued at the Jet Propulsion Laboratory during PrepSKA⁴²³ and beyond.

Beginning with a need to develop LNAs for the Large Adaptive Reflector (LAR),⁴²⁴ Leonid Belostotski at the University of Calgary developed a series of LNAs for PAFs in general and many other radio telescopes (e.g., (Belostotski & Haslett, 2006)).

At radio frequencies less than about 500 MHz, the situation is quite different because noise from the sky, itself, is a major contributor to overall system noise for the telescope. In this situation, uncooled LNAs are competitive. For SKA1-Low, which is designed with thousands of elemental antennas (see Sect. 6.5.3), LNAs can be made small enough to be directly attached to each antenna.

6.6.1.1 Wide Band Single Pixel Feeds (WBSPFs)

WBSPFs were included in the 2013 design baseline⁴²⁵ for only two bands, Band 1 (350–1050 MHz) and Band 5 (4.6 13.8 GHz) (Tan et al., 2016). Band 5 was later converted to two octave bands for reasons outlined below.

WBSPFs were a featured area of SKA innovation and research long before the PrepSKA era. They are feed antennas used at the foci of dishes to receive concentrated emission from the dish optics and are considered ‘wide-band’ if the ratio of the upper to lower frequencies is greater than two (an octave). This is in contrast with traditional octave band feeds (Granet et al., 2008), for which this ratio is about two. The design goal is to produce a beam that has a constant diameter over the entire frequency range, whereas the natural tendency of antennas is that the beamwidth scales with wavelength.

The ATA pioneered the use of a novel cooled WBSPF (0.5–11 GHz), consisting of a log-periodic dipole antenna encased in a glass cryostat so that the entire feed and Low Noise Amplifier was cooled to about 80 K⁴²⁶ (see Sect. 6.4.2). They were considered an important innovation in SKA Memo 100 (see Sect. 6.2.1.4), the first comprehensive set of SKA specifications. Germán Cortés-Medellín summarised the performance of eight different WBSPF designs being put forward for the SKA at the DVA-1 CoDR meeting in Socorro⁴²⁷ in 2011.

The appeal of WBSPFs was obvious. Only one or two feeds are required instead of several, the cost is potentially much lower, and it was possible that only one set of RF electronics (the signal chain) would be needed. However, difficulties arise because the beam is usually only quasi-constant with frequency. This affects the efficiency, imaging dynamic range and the spillover noise (from the feed beam spilling over the edge of the dish and intersecting the ‘hot’ ground) (see Sect. 6.4.3.3). The pros and cons of WBSPFs were summarised in a presentation⁴²⁸ at the PrepSKA WP2 meeting in Manchester in 2011.

The panel report from the System CoDR review⁴²⁹ in 2010 suggested that the SPDO “Plan a roadmap of the introduction of innovative technologies which will become available in later phases (e.g., WBSPF)”. WBSPFs later became part of the AIP (see Sect. 6.2.2.9). Post-2012 developments in WBSPFs can be found in SKASUP6-24.⁴³⁰

6.6.2 Signal Transport

An important example of enabling technology diffusion for the SKA is the rapid development of data transport by optical fibre (see point {H} in Chap. 6 introduction), a core part of the ICT ‘revolution’. By 1995, when the SKA concept was already circulating in the radio astronomy community, a frequent measure of performance, capacity-distance per fibre had doubled every 12 months since 1975 (Agrawa et al., 2016), faster than Moore’s Law (see point {G} in Chap. 6 introduction). A high rate of capacity growth has continued to the time of writing. Except for cost, in 1995 this would have made feasible data-rates of about 300 Gigabits per second (Gpbs) transmitted from antenna stations at 3000 km, more than enough for the SKA. By 2006 in Europe, VLBI stations were routinely connected by optical fibre⁴³¹ and the use of optical fibre in radio telescopes had become widespread (McCool et al., 2006).

However, cost is still a dominant factor for the SKA. But rather than the fibres themselves, the cost of trenching was the largest factor. This led to a major mathematical study of the most efficient network of trenches in a dense array of antennas (see Fig. 6.7).⁴³² Signal transport costs over long distances led South Africa to propose in their site bid of 2011 to co-locate the SKA correlator-beamformer and the processing super-computer to an “Astronomy Complex” near the site of the telescope, itself.⁴³³

Although the fibre optic cable is not a major part of the cost, the circuitry at the endpoints, including the cost of digitising the RF signals from the antenna, is expensive and requires considerable power. Although for dishes with PAFs or WBSPFs, the total bandwidth is higher than for typical single-pixel feeds, it is still well within reach of commercially available optical fibre systems. However, for the original configuration of the SKA-Low telescope discussed in Sect. 6.5.4, there would have been as many as 10,000 antennas in a station. This was later reduced to 256 antennas in more stations. In either case, the cost and practicality of digitising the signals from each individual antenna would have been prohibitive.

A much less expensive approach was to send RF signals directly over optical fibre, so called RF-over-fibre (RFoF). In its simplest form this technique uses an amplified version of the electrical RF signal from an antenna to modulate the amplitude of a laser whose output is connected to a fibre. At a central receiving end, a photodetector is used to recover the electrical RF signal, which is then digitised before being transmitted to a correlator or beamformer. Apart from being less expensive, the advantages are that devices are smaller, and less power is required. RFoF data-transport was studied extensively⁴³⁴ and evaluated in SKA-related prototypes: the Northern Cross Telescope,⁴³⁵ the Karoo Array Telescope (KAT),⁴³⁶ and ASKAP.⁴³⁷ Although the many pros and cons of this approach are beyond the scope of this discussion, RFoF is now part of the design of SKA1-Low, under construction at the time of writing.

Signal Transport provides the connective tissue for a geographically diverse project like the SKA. Because the system architecture was still quite immature at the time of the Signal Transport and Networks Concept Design Review (CoDR) in 2011, it was clearly difficult to provide a mature Signal Transport architecture. Scaling studies⁴³⁸ were used to partly overcome these issues, but nevertheless the risks of using RFoF continued to be highlighted in the panel report.⁴³⁹

6.6.3 Astrophysical Transients, Pulsar Searches and Timing

Pulsar astronomy became a veritable industry after their discovery in 1968. A series of subsequent discoveries have clearly illustrated their importance to astronomy and to fundamental physics (see Chap. 5). Most observations and searches have been carried out with large single dishes, notably the Parkes radio telescope in Australia, the Lovell radio telescope at Jodrell Bank in the UK, the Effelsberg radio telescope in Germany, and the Green Bank and Arecibo Telescopes in the USA. The Handbook of Pulsar Astronomy (Lorimer & Kramer, 2004) contains a synopsis of pulsar astronomy as of 2004 and a very useful overview of basic instrumentation for pulsar observations.

Translating the science goals of pulsar research (see Chap. 5) into a set of potential SKA observational programs provides a way of ascertaining design requirements: (1) Assemble a complete census of Galactic pulsars. (2) Use a sub-set of very stable pulsars (a pulsar timing array), to detect and characterise nano-Hz gravitational waves. (3) Use a sub-set of pulsars in binary systems, especially with black-hole companions, as tests of General Relativity in extreme environments. Additionally, search for a pulsar orbiting the supermassive black hole at the centre of the Galaxy. (4) Use pulsars as probes of the Galactic Interstellar Medium, the plasma that permeates the Milky Way. More generally keep discovery space in the time domain as wide-open as possible for discovery of unknown phenomena.

Array telescopes before the SKA were not normally designed to observe fast time-domain phenomena.⁴⁴⁰ The need to detect and observe astrophysical transients was foreseen in a memo from the early 2000s,⁴⁴¹ in which detailed science and important instrumental requirements were described. In SKA Memo 86⁴⁴² and SKA Memo 97,⁴⁴³ Jim Cordes developed modified definitions of survey speed and survey completeness to consider the nature of transient sources, pointing out that the time domain is a relatively unexplored “axis of discovery”.

As shown in Fig. 6.33, when a transient signal of astrophysical origin traverses the path through the ionised medium from the source to the telescope, it undergoes dispersion, which means that the signal is ‘stretched’ over frequency so that the high frequency component arrives sooner than the low frequency component. This causes an intrinsically narrow emitted pulse, which because of its narrowness covers a wide band of frequencies, to follow a parabolic curve downward towards lower frequencies with time. If the receiver band is divided into narrow frequency channels, a display of frequency versus time, the dynamic spectrum, will show the parabolic track.

Fig. 6.33

Left: The time-frequency (dynamic spectrum) signature of a dispersed pulsar pulse or an astronomical transient. The red solid curve suggests trials of unsuccessful lower dispersion-measures until the correct value has been reached. Right: Time-alignment after a matched de-dispersion operation has been applied. Bottom: The red plot shows the increase in signal-to-noise of the pulse after the de-dispersion operation and a summation over frequency has been applied

If the transient is a one-off event, it must be strong enough that it can be detected in a narrow frequency band. If the transients are pulses from a pulsar, then the train of pulses can be detected even if the pulses are weaker than the telescope noise because of their precise repetition. Hence, when searching the sky for weak unknown pulsars, the telescope must be capable of searching a space consisting of the unknown repetition rate and the unknown dispersion. This is equivalent to searching for all possible tracks in the large space shown in Fig. 6.33, large because pulsar repetition periods cover a range from about 1 ms to 10 s, and measures of dispersion cover a similarly large range.

There are two aspects of telescope architecture that are directly related to time-domain observations, especially searching for pulsars. The first is a beamforming capability which is implemented in the correlator-beamformer (CBF) as described in Sect. 6.6.4. The figure in SKASUP6-26⁴⁴⁴ shows the CBF outputting streams of data for each beam to be processed. For example, in the final design for SKA1-Mid, there are about 1500 beams, resulting a huge data rate, all to be searched in parallel. This is carried out by a pulsar search engine, a specialised computing device, designed to search the period-dispersion space described above (see the figure in SKASUP6-26). This device produces a small number of pulsar candidates that are further analysed in off-line software.⁴⁴⁵

The second is the array configuration. The ideal arrangement for time-domain observations is for all the collecting area to be concentrated at the centre of the array, the core. This is essentially equivalent to a large single dish, which provides the maximum sensitivity and the largest array beam (i.e., search area on the sky). The many competing goals for the design of the array configuration are discussed in Sect. 6.2.2.10 and depicted in Fig. 6.8. The compromise configuration, shown in the Box in SKASUP6-6,⁴⁴⁶ contains a dense core, partly to satisfy pulsar search requirements. With this configuration there is an optimum area around the core to use for pulsar searching. For example, the Baseline Design document⁴⁴⁷ provided an estimate of approximately the inner half of the SKA1-Mid antennas to be used for pulsar searching.

Pulsars have an emission spectrum that is stronger at low frequencies than high ones. But dispersion is also stronger at low frequencies. In general, the tension between these two effects leads to a broad optimum frequency range for pulsar searching, around 1 GHz. This means that the telescope capabilities at this frequency are very important for pulsar astronomy, particularly because some of this frequency range is badly contaminated with RFI.

Capturing rare transients spawns additional design requirements. If a strong dispersed signal is detected in one of the pulsar engines, it can be captured in the pulsar engine’s data store, but this will not provide much information on its direction. However, if output data from each antenna is stored for a short period of time in a data buffer before the event, then an image can be formed from these data to try to determine the position of the source on the sky. The contents of this ‘ring buffer’ is captured upon the detection of a transient and saved for subsequent analysis.

Discovering new pulsars is just a step towards the ‘real science’, which is enabled by follow-up observations to precisely track their Times-of-Arrival (ToAs). Strong-field tests of gravity and the detection of long-period gravitational waves require follow-up of a pulsar sub-set with both the capability to detect acceleration of pulsars orbiting compact objects and a long-term timing program. Only a sub-set of very stable pulsars or those that indicate accelerations will be covered in a long-term timing program. In instrumental terms, this means that measuring ToAs of pulses must be ‘time-tagged’ with a precision of a about 10 nanoseconds, traceable over a period of about 10 years, a ratio of about 10¹⁶. Such precision timing will reduce instrumental effects to the point where ToA scatter will be dominated by intrinsic scatter in the pulsar signals, themselves. Timing is carried out by a pulsar timing engine (see the figure in SKASUP6-26⁴⁴⁸) in conjunction with an Observatory ‘clock’⁴⁴⁹ that maintains accurate timing.

The foregoing illustrates the design complexity required to enable exploration of the radio-astronomy time-domain. As described above for the array configuration, compromises with other design goals and with cost inevitably had to be made. Based on the system architecture described in SKA Memo 130⁴⁵⁰ and modelling the results of an earlier survey using the Parkes telescope, estimates of computational requirements and processing options⁴⁵¹ were made. One driving aspect stood out: finding highly accelerated pulsars in a tight orbit around a massive object requires about 100 times the processing power than that required for isolated pulsars. This work led to a more complete high-level description⁴⁵² of pulsar search processing written for the Central Signal Processing (CSP) CoDR in April 2011.

Another aspect of this work was optimising a survey of the entire visible sky from the two SKA telescope sites. While frequencies covered by SKA1-Mid are best for regions near the Galactic disc where the interstellar medium (ISM) is most dense (i.e., high dispersion), a suitably equipped SKA1-Low could be used to survey the rest of the sky, where the ISM density falls off (low dispersion). For this and other reasons, both telescopes contain similar array beamformers and pulsar engines.

Over the years, pulsar astronomers had already developed sophisticated processors, similar in many respects to those needed for the SKA. By 2004 (Lorimer & Kramer, 2004), most of the fundamental algorithms had been developed. Further progress could only be made by developing optimised architectures for implementing them in the available hardware of the day. From 2003 to 2007, the field moved away from recording the signal from the telescope and processing off-line to processing the signal in real time at the telescope site. By 2010, many implementations of real-time processors were being used, and versatile computer code (van Straten & Bailes, 2010) was widely distributed.

In principle, just replicating these devices would suffice for the SKA, since it is just a matter of one device per beam. As with the correlator-beamformer case (see Sect. 6.6.4.1), there was a tendency to jump directly to design solutions based on bespoke hardware.⁴⁵³ But the field took another direction when Graphics Processing Units (GPUs) became widely available. They can very efficiently execute Fourier transforms and other basic functions, the key bottleneck in pulsar processing (e.g., see (Barsdell et al., 2012)). Unlike other components of the SKA system, it does not make sense to build too early due to the continuing rapid development of electronic technology (see point {G} in Chap. 6 introduction). Hence, continuing to build and develop pulsar processors (search and timing) using the latest technology on various available telescopes (e.g., (Bailes et al., 2016)) will benefit the SKA until finally it becomes necessary to build specifically for the SKA telescopes (Stappers et al., 2018). Further developments in this story are beyond the scope of this book.

6.6.4 Correlators and Beamformers

The correlator, often called the ‘processing heart’ of a radio telescope, is a central system that receives signals from antennas and processes them to produce viable scientific data. Modern correlators are based on the so-called FX architecture.⁴⁵⁴ SKASUP6-26⁴⁵⁵ explains how correlators and beamformers function in the SKA architecture. Combined to make the correlator-beamformer (CBF), they are specialised super-computers designed to carry out simple processing steps on vast quantities of data. The output data produced by the correlator represents points in the u-v plane for each channel (see Sect. 6.2.2.10 and the Box in SKASUP6-6⁴⁵⁶). From the data in the u-v planes, it is possible to create detailed radio images over the field-of-view, using the methods outlined in Sect. 6.6.5. Beamformers produce data streams by combining the data from the individual antennas in such a way that the output resembles that from a single large antenna with a narrow beam on the sky.

The Signal Processing CoDR in April 2011 was an opportunity for institutes to present their wares and to illustrate their capabilities. In a high-level description,⁴⁵⁷ which was virtually a case study in a system-engineering approach to an early design review, the scene was set by Wallace Turner, SPDO’s domain specialist in signal processing. It covered the system-engineering gamut: motivations, requirements, known algorithms, risk, costs, technology roadmap, and strategy to proceed to the next phase.

Ten architecture options for correlators from seven institutes, three beamformer options, and five options for pulsar processing were presented at the Signal Processing CoDR.⁴⁵⁸ This became known as the Battle of the Boards (see next section). As highlighted in the report of the review committee,⁴⁵⁹ the issues were how to compare the various options on the same basis: how to narrow them down without carrying forward too many and eliminating some too early, and how to make choices in the light of technology advances. The backdrop was the rapid changes in technology epitomised by Moore’s Law (see point {H} in the Chap. 6 introduction). This was anticipated in a technology roadmap⁴⁶⁰ produced at the time but turned out to be difficult to maintain in the ensuing years.

These issues were only resolved during the pre-construction period after 2012, when the Frequency Slice Processor was introduced.⁴⁶¹ By that time, many of the competing institutes had lost interest or moved on to other projects. It is interesting to note that resolution of the options dilemma was not achieved through attempting to choose options strictly on a technical or system-engineering basis, but rather through an allocation process among participating countries. While the process was messy (see point {E} in the Chap. 6 introduction), it had the overall effect of choosing the institutes with the strongest motivation, resources, and best track-records in the field. While this did not completely avoid competition and challenges, it provided a simpler basis for moving forward.

6.6.4.1 Battle of the Boards

Throughout the PrepSKA era, bespoke digital hardware was the favoured solution. This was because software correlators, based on commercially available computers, even supercomputers, could not handle the data volume. Each major radio-astronomy institute wanted to develop its own solution to this problem, mainly for telescopes for which they had already developed equipment, including the SKA pathfinder and precursor telescopes (e.g., (Szomoru, 2011; Hampson et al., 2014)). An overview of these options was presented by Brent Carlson at the 2010 PrepSKA WP2 meeting in Manchester.⁴⁶² In most cases, the processing hardware was based on Field-Programmable Gate Arrays (FPGAs), which are commercially available integrated circuits that can be configured to efficiently process data at high speed. Like other electronic devices, FPGAs have rapidly improved and provide a good route to building data-processing systems on a small to medium scale, rather than the very expensive development of ‘hard silicon’.

During the PrepSKA period, many institutes proclaimed the utility of their boards for all manner of uses, including for the SKA, itself. The most successful of these in this sense was the Collaboration for Astronomy Signal Processing and Electronics Research (CASPER), headquartered at the University of California at Berkley (Werthimer, 2011). CASPER set out to produce designs of reconfigurable boards that could satisfy the needs of many institutions for a variety of purposes: correlators, beamformers, pulsar search and timing machines, etc. The project received a big boost when the MeerKAT project decided to join the collaboration and on behalf of the collaboration, designed what became known as the ROACH board, which was then used for the MeerKAT digital backend.⁴⁶³ This board found uses in many different telescopes. The CASPER hardware was not necessarily better than the other boards, but the collaboration was specifically formed to pool resources (especially shared software/firmware) and was promoted heavily in the US.

At low frequencies, for which bandwidths are smaller, software correlators are viable instead of customised hardware. The LOFAR project did build a large software correlator in 2010, based on an IBM Blue Gene/P computer (Romein et al., 2010). This was replaced in 2018 by a GPU-based⁴⁶⁴ correlator (Broekema et al., 2018), which uses commercially available hardware (see above {G} in Chap. 6 introduction). Although the software correlator for LOFAR was only marginally successful, software correlators have been successful for processing VLBI observations (e.g., (Deller et al., 2011)).

The subsequent evolution of SKA correlator architectures post-2012 is discussed briefly in SKASUP6-27.⁴⁶⁵

6.6.5 Radio Images: SKA’s Ambitious Software Requirements

Information processing has always been a constraint on the theoretical performance of radio telescopes. Intuitively, the information available to a radio telescope is proportional to the number of radio-imaging pixels in the field-of-view of the telescope times the number of frequency channels to be acquired, often referred to as a 3-D image cube. However, the signals that convey this information are very weak and usually masked by noise. Moreover, a practical telescope does not capture the signals perfectly and contains errors. Finally, RFI signals (see Sect. 6.2.2.12) have the effect of making some frequency channels unusable. The software component of radio telescopes has the task of forming the image cubes from the correlator output data, applying calibrations to reduce the effects of errors, and ameliorating the effects of RFI and noise.

Calibration is a measurement or procedure that reverses the effects of errors in the telescope. There are many types of calibrations, but many require software-intensive processes. Although errors do not necessarily arise from the software itself, residual errors remaining after the data has been calibrated limit the performance of telescopes in general. An important additional effect occurs at frequencies covered by SKA-Low. The ionosphere inserts a time-varying distorting screen in front of the radio sky that, in its simplest form, moves the apparent position of radio sources on both short and long timescales. This effect must also be removed in software.

For example, a key science objective of the SKA was to be able to observe the very weakest radio sources, mainly those expected to be from the earliest galaxies in the 21-cm wavelength window. This capability, high dynamic range imaging, requires the detection of sources that are about 10 million times weaker than nearby strong radio sources in the sky (see Sect. 6.4.3.2). Even the smallest of errors will affect this goal.

Many software packages devoted to imaging and calibration have been developed over the decades of radio interferometry. A mathematical formalism for radio interferometry telescopes, the measurement equation, was formulated by Johan Hamaker and his collaborators in 1996 (Hamaker et al., 1996), although the principles were known long before. Ever since then, important practical advances in algorithm development at various institutions (e.g., (Rau et al., 2009)) have been made, many of which have been relevant for developing the SKA software.

Historically, telescope operators have intentionally throttled the dataflow from the correlators to the downstream computers (ingest rate) so that they can keep up with the calculations or data storage. This has not been a major problem in the observer community because many are interested only in narrow objectives in field-size or frequency channels. Also, observers frequently carry out data reduction using their own compute facilities, and don’t wish to be burdened with too much data. Nevertheless, these practices are discarding discovery space (see also Sect. 6.2.2.8). Because of scientific interest in re-examining archival data, it is important not to discard any data in the future. This was recognised as important in the PrepSKA era but not given high priority at that time.

The SKA planned to carry out full-field imaging continuously, and even to support very wide-field imaging by stitching adjacent images together or by scanning across large areas of sky. Storing the raw interferometer data was recognised as being very difficult because of limited data-storage capabilities, forcing imaging and other data reduction to be carried out in quasi-real time. Hence the observing schedule, data buffers, and data-processing computers must be jointly managed so that computing can keep up with the pace of observations. More importantly, raw correlator data will be eventually discarded, and observers cannot go back, for example, to reapply some calibrations. This is a break from tradition for major array telescopes for which observers have access to original data.

Computing capability has improved in the intervening years, but at the time of writing, the SKA will still not be capable of storing the raw data.

Of the many design aspects to fill in after the formulation of the Reference Design in 2006 (Sect. 6.2.1.3), one was the software capabilities required. In an early SKA memo on dish diameter,⁴⁶⁶ for example, Tim Cornwell argued that the computing costs of wide-field imaging scaled as d⁻⁸, where d is the dish diameter⁴⁶⁷ (see Sect. 6.4.3.1). Nevertheless, in 2005 he advocated taking the Large-Number—Small-Dish (LNSD) design to an extreme with 1.5–3-m diameter dishes.⁴⁶⁸ But to achieve this, a separate beam-forming stage would have been required to narrow the beam to a manageable field-size. In the end, issues other than software were more influential in determining dish size (see Sect. 6.4.3.1), recognising that the SKA could ‘grow into’ more capable software with time, but dish size would be immutable. Similar discussions were held around the size of Aperture Array stations.

As noted in Sect. 6.2.1.3, one of the PrepSKA responsibilities taken on by the US Technology Development Program (TDP)⁴⁶⁹ was to carry out investigations of and to act as a clearing house for areas of high risk associated with SKA-Mid: dish designs, and calibration and processing software. A Calibration and Software Group (CPG) was formed, chaired by Athol Kemball, and consisted of about 12 members brought together key persons interested in this aspect of the SKA.⁴⁷⁰ They produced a series of documents which directly addressed some of the issues described above: high dynamic range imaging,⁴⁷¹ peta-scale computing,⁴⁷² implications for antenna and feed design for wide-field imaging,⁴⁷³ and the computation implications of realistic antenna arrays.⁴⁷⁴

During the PrepSKA period and continuing afterwards, the Calibration and Imaging CALIM⁴⁷⁵ meetings were important fora for specialists in imaging with radio interferometers to discuss approaches to solving problems for the SKA and other telescopes. Staging the meetings across the globe⁴⁷⁶ increased the visibility and importance of these meetings. The CALIM meetings were widely attended and provided the spark for many advances in the development of algorithms for radio telescope imaging. Sponsors were ASTRON, NRAO, SARAO, CSIRO and UWA.

Three significant programs were started in the PrepSKA period to develop software tools to improve the efficacy of calibration:

• MeqTrees (Noordam & Smirnov, 2010), a simulation tool that implements the measurement equation. This tool is particularly useful in simulation and calibration of so-called direction-dependent effects, manifestations of the complex behaviour of real antenna beams rather than idealised ones. It has played a major role in improving the imaging dynamic range of radio telescopes.⁴⁷⁷

• OSKAR (Mort et al., 2010), a software simulator which was designed to simulate beamforming for aperture array telescopes, specifically SKA-low and at an earlier stage, dense aperture arrays (see Sect. 6.5.5). A version is still in use at the time of writing.⁴⁷⁸

• CyberSKA.⁴⁷⁹ This was a distributed computing infrastructure funded by Canada’s National Research and Education Network (NREN) as a precursor to an SKA Regional Centre (SRC) in Canada. It was used to aggregate and process large datasets to which Canadian astronomers already had access. In subsequent years, the SRC concept has expanded to a global network of regional computer centres, through which observers will access and sometimes process SKA data products. CyberSKA was a very early experiment, especially since SRCs in the PrepSKA period were not much more than notional.

6.6.5.1 Computer Engineering

Computer engineering is the implementation side of radio astronomy imaging. As noted above, the problem requires super-computing scale and unique software. Comments in 2003 by Marco de Vos, representing Software Engineering in the IEMT in a report to the ISSC,⁴⁸⁰ bluntly outlined the challenges.

These challenges were the impetus for the Convergent Radio Astronomy Demonstrator CONRAD collaboration^481,482 between the ASKAP and MeerKAT groups in Australia and South Africa, respectively.⁴⁸³

Technically, the computer processing algorithms needed for the SKA are highly ‘parallelisable’. This means that the computer architecture can provide the means to direct data to processors that independently carry out a slice of the processing. While it is beyond the scope of this book to explain this in detail, this fact greatly improves the prospects for computational feasibility for the SKA.

Software development was recognised as one of the greatest challenges in the period up to 2012 and remains so at the time of writing. Estimates in 2004 were that 1000–2000 person years of effort would be needed.⁴⁸⁴ An heuristic scaling analysis concluded that the scientific and operational requirements must be scaled back, and at least 20% of the budget should be allocated to software development.⁴⁸⁵ This document pointed out that the level of software effort scales with size of the project raised to a power greater than one (i.e., the so-called diseconomy of scale).

The computing cost scaling was re-estimated in 2005, where the scaling was closely related to the size of the field-of-view.⁴⁸⁶ Again, a recommendation to limit the size of wide-field images resulted. By 2010, it appeared that peta-scale computers would not be sufficient for the SKA, that exa-scale computers would be needed, and that scaling of existing algorithms and code would not work.⁴⁸⁷ These issues remain as the SKA is under construction, but there is a much better understanding of how to tackle them. One advantage of software and computing development is that the telescope can begin observations long before the ‘ultimate’ software package is available.

The Software and Computing (S&C) Concept Design Review (CoDR) was organised in February 2012,⁴⁸⁸ by Duncan Hall, the S&C domain specialist. It was the last of the sub-system CoDRs and covered software for monitor and control, time-domain processing (see Sect. 6.6.3) and imaging (visibility processing). An analysis of the science requirements^489,490and of the visibility processing⁴⁹¹ revealed that the SKA’s ambitious science goals would indeed push software and computing to the limits of (then current) technology, especially because of the quasi-real time processing requirement described above. The software engineering analysis⁴⁹² outlined the new challenges presented by the SKA: novel scale, very large data flow and processing, management across the globe, small number of experienced people, novel front ends (if adopted, e.g., PAFs, dense AAs, WBSPFs) and substantial algorithm development.

The S&C panel report⁴⁹³ noted a maturity lag in the S&C development compared with other parts of the SKA design and compared, for example, with the Large Synoptic Survey Telescope (now the Rubin Observatory) at a similar stage. The answers to all the standard CoDR questions were sceptical. They recommended strong centralised management and recruitment of a full complement of software staff to offset the risk.