"#$!%&'()$!*+,-.$/)$!0))(1!

Transcription

1 $ "#$%&'()$*+,-.$/)$0))(1 02/)-3454 "#$%&$'()*+,$-./)+.-$ &0$1*23$&445$ $ $ 6-(.7)+$894$ 6-7/(8/0'/#-)9 1:;-.$<)(,-.$ <=:7(>$?9@9$@AB$<)+.)(C7*;$ D40$&EEF4D4G$ <)(+-22$?+7H-(.7C3$ I)(,-.J:.C()9I)(+-229-,*$ 0'/#-)2(7:;()/+8+<(7/29?9@9$@AB$<)+.)(C7*;K$88$#+.C7C*C7)+.$$L=CC/KMM*..N:I9:.C()9I)(+-229-,*M;-;O-(.MP$ #+/*C$Q();$#+C-(+:C7)+:2$@AB$R:(C+-(.$ $

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i

3 Table of Contents Executive summary 1 I. Science overview 2 II. Technical Implementation 7 II.1 Antenna array description Antenna Array Characteristics (Table 1) 9 II.2 Observation strategy 19 II.3 Facility construction 21 II.4 Instrumentation: II.4.1 Optimized Reflectors 24 II.4.2 Wideband Single-pixel Feed Antennas 27 II.4.3 Central Data Processor 30 II.4.4 Phased array feeds 33 II.4.5 Aperture Array 36 III. Enabling Technology 39 IV. Facility and Science Operations 41 V. Programmatics and Schedule V.1 Current Organization 45 V.2 Program Risks and Risk Mitigation 46 V.3 Overall schedule 48 V.4 Key Phase Duration Schedule 50 V.5 Key Event Dates Table 50 V.6 Schedule Charts 51 VI. Cost Section 53 VII. Changes since previous NRC recommendation 70 ii

4 "#$%&'(#)*%++,-. The Square Kilometre Array (SKA) is a revolutionary telescope program that will address a broad range of key science areas in galaxy evolution and cosmology, fundamental physics, and astrobiology. It will also be a superb discovery instrument. The key science goals of the SKA are in place and a well-structured, global technology development program funded at substantial levels is under way. While construction is not funding-ready in 2010, previous and ongoing technology development and cost estimates will lead to technology choices and definitive costings early in the decade and construction proposals for phased deployment on that time scale. These efforts are matched by planning efforts for governance and funding of the project that also are organized by the SPDO (SKA Program Development Office in Manchester, UK). The emphasis of the activity described here is on the mid-frequency array (SKA-mid) because the science goals are timely and the world-wide community is working toward bringing the technical readiness level to that needed for construction by the middle of next decade. Phase 1 comprising low-and-mid frequency arrays will be funding-ready in ~2013 and Phase 2 to complete the array is planned to begin in ~2017. The SKA is presented to Astro2010 as warranting high priority in 2010 on the basis of the science case, the level of global interest and commitment at both of the scientific and governmental levels, the comprehensive international technology development program now in place, and the existence of a well managed international SKA organization to execute the program. It is recognized that funding decisions will depend on the delivery of technology-ready, risk-mitigated proposals along with successful completion of milestones. The submission here includes indicative technologies, costs, and timescales. These will be refined by the SPDO using input from both the international scientific community (via the SKA Science and Engineering Committee, SSEC) and funding agencies (via the Agencies SKA Group) as the SKA program progresses. The SKA presently involves 55 institutions in 19 countries, with the U.S. as a one-third partner. Activities within the U.S. are organized through the U.S. SKA Consortium that includes 11 organizations, including universities, laboratories, and the two national radio observatories (NAIC and NRAO). The Consortium is conducting an SKA Technology Development Project, funded by the NSF in accordance with the last decadal-survey recommendation for the SKA, which contributes key expertise and technology to the global design effort. The baseline design for "#$%&'( )*&+,'-.- /000 12%& reflectors separated by up to a few thousand kilometers. Signal analysis will provide the unprecedented sensitivity, survey speed, and resolution needed for the key science. Two viable sites for the SKA core array have been identified (Western Australia and South Africa), on which precursor arrays will further develop some of the technologies needed for the SKA. Project milestones over the next few years include site selection targeted for 2012, technology selection, and critical design reviews that will lead to the first phase of construction in ~2015. The anticipated 1/3 contribution of the U.S. is (2009 dollars) ~$725M for construction, $65M/yr for operations, and $10M/yr for programmatic support of U.S. scientists, including a U.S. based SKA Science Center. The draft spending profile for the U.S. starts at about $5M/yr in 2012 and 2013, reaches about $100M/yr during peak construction starting late in the decade and continues near this level for full SKA operations starting in the following decade. 1

5 I. "#$%"$&'($)(#$*& The SKA science case has developed over the last 10 years starting initially from a grass-roots realization that a large increase in sensitivity is needed at radio wavelengths and in the last six years through a deliberate process conducted by the SKA Science Working Group (SWG). Like all major science facilities, the SKA will conduct a far broader range of observations than the key science areas referred to here, as may be seen in Science with the SKA (Carilli and Rawlings, eds., New Astronomy Reviews, 2004), articles of which may be found at The SWG under the leadership of the SKA Project Scientist continues to examine and revise the science case in light of engineering/science tradeoffs and, of course, new developments in astronomy and astrophysics. Q1. Describe the measurements required to fulfill the scientific objectives expected to be achieved by your activity, providing up to four examples of the science this project is designed to address. In our responses to RFI#1 and to supplementary questions posed by the RMS committee, we identified four key science areas for the SKA-mid, which we focus on here as exemplars for the kinds of observations that need to be done. Key Science Goal Measurement/Observation Inventory the atomic gas (HI) content of the Universe over at least the last half of the Universe s age Deep HI field(s) to z >~ 2 1. Galaxy Evolution, Cosmology, & Dark Energy 2. Strong Field Tests of Gravity Using Pulsars & Black Holes 3. Origin and Evolution of Cosmic Magnetism 4. The Dynamic Radio Sky Deep continuum field(s), detecting star forming galaxies to z ~ 2 and potentially to z ~ 7 Track star formation and AGN activity/feedback over a significant fraction of cosmic time Image CO emission from galaxies at z ~ Constrain the evolution of dark energy via baryon acoustic oscillations (BAO) Detect and study gravitational waves from inspiraling supermassive black hole binaries Test theories of gravity using ultra-relativistic binaries Probe spacetime environment and dark matter in Galactic center Track the evolution of magnetic fields in clusters, galaxies over cosmic time; assess roles of primordial magnetic fields and dynamo proccesses Discovery and census of fast and slow radio transients (luminosities and event rates) 10 All-hemisphere survey of HI emission to z > 1 All-hemisphere survey for stable millisecond pulsars Timing observations of stable millisecond pulsars to form pulsar timing array (PTA) Galactic plane and globular cluster survey for ultra-relativistic binaries Timing observations of ultra-relativistic binaries Survey Galactic center for pulsars orbiting Sgr A* Timing observations of pulsars orbiting Sgr A* All-hemisphere survey to construct Faraday Rotation Measure grid Deep continuum field(s) toward clusters of galaxies All-sky surveys Targeted fields toward known or suspected locations of transients 2

6 Q2. Describe the technical implementation you have selected, and how it performs the required measurements. The baseline design for the SKA-mid is an interferometric array of parabolic dish antennas, initially outfitted with a wide-bandwidth feed and receiver system (so-called wide-band single pixel feeds, WBSPFs). Possible expansion paths aimed at increasing the accessible field of view (FoV) that are the focus of world wide development include outfitting the antennas with phased array feeds (PAFs), which form multiple FoVs by sampling the electric field in the focal plane of the antenna; separate dense aperture arrays (AAs), which consist of electrically short antennas phased together to form effective apertures; or both. These subsystems are described in Section II/Instrumentation. This baseline design has resulted from a series of technology selection decisions by the SKA community over the past decade. The required measurements (viz. Question #1) can be grouped into two broad classes, imaging and non-imaging. In this respect, the SKA-mid will function similarly to many of the existing radio astronomical arrays, including the VLA, VLBA, ATA, WSRT, GMRT, and ATCA, and planned arrays such as ASKAP and MeerKAT. Imaging: In this mode, the electrical signals from the antennas in the array are time delayed to account for the geometry of the array, then all unique pairs of antennas are multiplied together. The result is a sampling of the Fourier transform of the image of the sky. The SKA-mid will perform the required imaging measurements by virtue of: Sensitivity: Determined by the total effective area and system temperature. Frequency Coverage: Determined by the frequency ranges of the WBSPFs and the PAFs. Angular resolution: Related to the maximum separation between the antennas. Survey Speed: Determined by a combination of the number of antennas, their FoV (i.e., whether outfitted with WBSPFs or PAFs), and the instantaneous processed bandwidth. Survey speed is defined as SS= (A e /T sys ) 2 where is the field of view (deg), A e is the effective area and T sys is system temperature. Note that etendue (A e ) often used in characterizing OIR telescopes is not a good measure for comparison between telescopes of vastly different wavelength bands. Non-Imaging: In this mode, the electrical signals from the antennas in the array are time delayed to account for the geometry of the array, then summed. The result is to treat the array as a single aperture. Many of the same considerations apply to the performance of the non-imaging modes as imaging modes, except that issues associated with antenna separations (baselines) are different. To minimize processing requirements, for example, compact configurations are favored for some non-imaging science, such as pulsar surveys. Q3. Of the required measurements, which are the most demanding? Why? The observations and measurements described in the answer to Question #1 place different demands on the telescope. 3

7 Measurement/Observation Demand/Requirement Justification Deep HI observations High sensitivity (A eff /T sys ) Faintness of HI line, coupled with System stability for long integrations high redshift Deep continuum observations High sensitivity (A eff /T sys ) High angular resolution System stability for long integrations High imaging dynamic range Faintness of typical star-forming galaxies, coupled with high redshift, and requirement to mitigate confusion Imaging high-z CO All-hemisphere survey of HI emission to z > 1 Pulsar surveys & timing Galactic center pulsar survey & timing Faraday Rotation Measure grid All Sky Transient Survey High sensitivity (A eff /T sys ) Operation to ~ 15 GHz High survey speed Automated survey imaging Robust data management High time resolution Robust data acquisition, management Operation 10 GHz High polarization purity Wide field of view, high time resolution 4 Faintness of typical star-forming galaxies, coupled with high redshift Faintness of HI line, coupled with volume of Universe to be surveyed, and large number of objects to be detected (~ 10 9 ) Intrinsic pulsar periods and size of parameter space to be searched Mitigate interstellar scattering Typical linear polarization of radio sources ~ 3-5% Full census on all time scales to the extent possible Q4. Present the performance requirements (e.g. spatial and spectral resolution, Strehl ratio, sensitivity, timing accuracy) and their relation to the science measurements. The following is based on the SKA-mid Design Reference Mission document, currently nearing its initial public release, and is a follow on to SKA Memo 100, Preliminary Specifications for the Square Kilometre Array. Parameter Requirement Observation / Measurement Sensitivity (A eff /T sys ) ~ 12,000 m 2 K 1 Deep continuum images Deep HI fields Pulsar surveys & timing Upper operational frequency 10 GHz nominal (~ 15 GHz desired) High-z CO images Pulsars near Sgr A* Survey Speed X -1 deg 2 m 4 2 All-hemisphere HI K survey in X years Maximum Baseline > 3000 km Deep continuum fields Temporal Resolution Polarization Purity Imaging Dynamic Range ~ 1 "s for timing ~ 50 "s for searching 40 db Pulsar surveys & timing, radio transient surveys Pulsar timing Faraday Rotation Measure grid > 70 db Deep continuum fields Science Case Galaxy Evolution, Large Scale Structure, & Dark Energy Strong Field Tests of Gravity Using Pulsars & Black Holes Galaxy Evolution, Large Scale Structure, & Dark Energy Strong Field Tests of Gravity Galaxy Evolution, Large Scale Structure, & Dark Energy Galaxy Evolution, Large Scale Structure, & Dark Energy Strong Field Tests of Gravity, Dynamic Radio Sky Origin & Evolution of Cosmic Magnetism Galaxy Evolution, Large Scale Structure, & Dark Energy

8 Q5. Present a brief flow down of science goals/requirements and explain why each instrument and the associated instrument performance are required. Each science goal (Question #1) typically implies multiple science requirements. We summarize the science requirements and focus on only the most demanding technical requirements. As for Question #4, much of this information is based on the SKA-mid Design Reference Mission and Memo 100. Science Goal Science Requirement Key Technical Requirement Inventory atomic gas content of Universe Image M * galaxies in HI at z ~ 2 Sensitivity ~ 12,000 m 2 K 1 Sensitivity ~ 12,000 m 2 K 1 Track star formation and Detect radio emission from LIRGs Maximum baseline > 3000 km AGN activity/feedback to z ~ 6 Imaging Dynamic Range > 70 db HI BAO measurement Detect M*galaxies in HI at z > 1 Survey Speed > deg 2 m 4 K 2 Gravitational wave Pulsar Timing Array Ultra-relativistic pulsar binaries Galactic center spacetime and dark matter content Evolution of cosmic magnetic fields Dynamic Radio Sky Survey sky for stable millisecond pulsars; long-term timing of an ensemble to better than 100 ns precision Survey Galactic plane for ultrarelativistic binaries; long-term timing of binaries Survey Galactic center for isolated pulsars and ultra-relativistic binary pulsars + long-term timing Faraday Rotation Measure grid with density of 5000 RM/degree 2 Comprehensive sky survey on fast and slow time scales Sensitivity ~ 12,000 m 2 K 1 Temporal Resolution ~1 µs (timing), 50 s (searching) Polarization Purity > 40 db Sensitivity ~ 12,000 m 2 K 1 Temporal Resolution < 50 s Polarization Purity > 40 db Sensitivity ~ 12,000 m 2 K 1 Temporal Resolution ~1, 50 s Operational Frequency " 10 GHz Polarization Purity > 40 db Polarization Purity > 40 db Wide field of view, wide range of time resolutions Q6. For each performance requirement, present as quantitatively as possible the sensitivity of your science goals to achieving the requirement. For example, if you fail to meet a key requirement, what will the impact be on achievement of your science objectives? The table below summarizes the impacts of not meeting the various technical requirements of the SKA-mid. We attempt to be as quantitative as possible in describing the observational impact or mitigation strategy. However, because interferometers degrade gracefully, we describe the scientific impact in more qualitative terms. 5

9 Requirement Sensitivity ~ 12,000 m 2 K 1 Upper frequency " 10 GHz Observation / Measurement Deep HI fields Deep continuum images High-z CO galaxies Pulsars near Sgr A* Survey Speed > X -1 deg 2 m 4 K 2 All-hemisphere HI survey in X years Baselines > 3000 km Temporal Resolution ~ 50 s Temporal Resolution ~ 1µs Polarization Purity 40 db Imaging Dynamic Range > 70 db Wide fields of view Deep continuum fields Pulsar surveys Precision timing Faraday Rotation Measure grid Pulsar timing Deep continuum fields Hemisphere HI survey; fast and slow transients Observational Impact of Descope Image more massive galaxies (in HI) and/or at lower redshift and/or use longer integrations Only image ULIRG or higher luminosity galaxies, at lower redshifts, or both Restricted to higher redshifts Detect pulsars only far in front of Sgr A* Only image galaxies with M(HI) > M * (HI), at z < 1, or both Reduced angular resolution Selection against fastest spinning pulsars Lower timing precision Lower density of RMs on sky, fewer lines of sight through galaxies and clusters Lower timing precision Only image ULIRG or higher luminosity galaxies, at lower redshifts, or both Survey incompleteness, less precision on dark energy Science Impact of Descope Fewer galaxies observed, at lower redshifts, and poorer constraints on gas properties of galaxies Fewer galaxies observed, at lower redshifts, and poorer constraints on star formation and AGN properties of galaxies Unable to detect first galaxies Unable to penetrate interstellar scattering, no pulsars found, and no constraints on Galactic center environment Reduced precision on dark energy measurements and poorer constraints on gas properties of galaxies Reduced ability to resolve morphology of galaxies to distinguish star formation and AGNs and increasing limitations from confusion Reduced ability to study gravitational waves and test General Relativity Reduced sensitivity to nano-hz gravitational wave perturbations Reduced ability to track cosmic magnetic field evolution Reduced ability to study gravitational waves and test General Relativity Fewer galaxies observed, at lower redshifts, and poorer constraints on star formation and AGN properties of galaxies Dark energy determination less competitive 6

10 "#$%&'()&*+#,-+%,%(.*.)/(#/0#.'%#123# II.1 (Answer to Q1): Overall description of the telescope, referring to Tables 1 & 2. SKA System Figure 1 is a system schematic showing the major elements of the SKA telescope. The SKA design uses aperture synthesis principles and consists of an array of antennas whose RF outputs are digitized and connected to a Central Data Processing (CDP) system. The CDP is a highly flexible real-time signal processor. Its primary purpose is to channelize the signals into frequency channels and to cross-correlate the signals from the antennas in pairs, so as to extract data in the Fourier transform (u-v) plane of the brightness distribution within the field-of-view defined by the main-beam solid angle of the individual antennas. This can be done simultaneously from all or multiple subsets of the antennas, depending on the range of spatial frequency coverage needed by the science program. In addition the CDP can carry out delaycorrected sums of the channelized signals. The summed signals essentially form telescope-beams on the sky, whose sensitivity is related to the entire collecting area used in the beam-sum. These signals will be further processed in real time to look for time-domain signatures, especially pulsar signals, but also other transient signals of astrophysical interest. The output of the CDP will be transmitted to a more general-purpose Science Computing Facility (SCF), where calibration of the data takes place, images of sky brightness are formed, and further analysis of time-domain effects are carried out. Algorithms for carrying out calibration and imaging are well developed for current dish-based aperture synthesis arrays, as are techniques for pulsar searching and timing. The volume rate of raw data from the SKA is too large to archive directly. Real-time image formation will be required. Images and other data products extracted from the images and time-domain data will be archived and distributed to SKA science centers around the world. Many of the science programs will be large, extended surveys. The survey science teams will each have specific requirements for real-time data processing and for handling/distributing data products. A key point for the SKA is that, by their very nature, aperture synthesis telescopes are flexible, and can be enhanced over time; the SKA is designed to be take advantage of this as much as possible. Competitive observations can begin soon after the number of installed antennas reaches 100 or so, similar to the collecting area of the Very Large Array. The rollout plan for the SKA will ensure that science begins as soon as possible. Table 1 (the requested array characteristics table) contains the SKA system parameters representing the current baseline design, not including supporting infrastructure. The subsystems are described in the following text and their cost is covered under the cost section of this document. In this document we break out five particular subsystems as instruments that require more detailed description. The first three are part of the end-to-end system: 1) Optimized reflector antennas. 2) Wideband single-pixel feeds and receivers. 3) Central data-processing system consisting of the correlator and a time-domain processor for pulsars and transients. These will become available when preliminary science operations begin and will develop according to the time plan presented in Section V. 7

11 Figure 1: A conceptual block diagram for the SKA, showing the interconnection of major components, the location of major parts and the flow of data from the antennas (top and left) to the signal processing facility, and off-site to a computer facility. Control interconnections between the Operations and Maintenance Center and the on-site components are not shown. In addition to the three subsystems/instruments of the baseline design, two additional field-ofview expansion subsystems are on a potential upgrade path to enable or enhance several of the survey science areas: 4) Phased-array feeds. 5) Dense Aperture Array. Antennas The antenna array for the SKA will comprise three thousand 15-m equivalent diameter offset- Gregorian dish systems, each equipped with two wide band feeds to provide overall frequency coverage of 0.3 to 10 GHz. Use of cryogenic receiver front-ends coupled to the feeds will maximize the sensitivity and survey speed of the array. A remote-controlled feed changer on each dish will also have provision for a phased array feed (PAF) to be installed as a future upgrade, to further enhance the transient survey capability of the SKA (see instrumentation section). Following the front-ends, the remaining sections of the receiver subsystem will provide gain, frequency conversion, digitization and optical modulation, such that the antenna system outputs will be digital bit streams on fiber-optic cables. Clock and local oscillator reference signals to the receiver subsystems will also be distributed via fiber optic cables. 8

12 Aperture Table 1: SKA Baseline System Description 3000, 15-m diameter parabolic antennas Lower Frequency 300 MHz Upper Frequency 10 GHz Antennas may be speced for 15 GHz Array Configuration Optimized for science & local conditions. Core (<1 km diameter) ~20% (600 ant.) Fractional total number of antennas Inner (<5 km diameter) ~30% (900 ant.) Mid (<200 km radius) ~25% (780 ant.) Outer (<3000 km radius) ~25% (720 ant.) In groups of 18 antennas (total of 40 stations) Antenna RF System Feed/LNA GHz Dual pol n (2 orthogonal), cyro-cooled LNA Feed/LNA GHz Dual pol n (2 orthogonal), cyro-cooled LNA/feed Core-to-Mid antennas Bandwidth * 4 GHz For each polarization RF Sub-bands 4 2 GHz sub-bands Outer Antennas (stations) Bandwidth * 4 GHz For each polarization RF Sub-bands 4 One output per station (digital beamformed) Digital Outputs Core-to-Mid antennas Sample streams * 4 Sampled 2 GHz sub-bands bits per sample 4 Outer Antennas (stations) Sample streams * 4 One summed beam, formed digitally per station bits per sample 4 Signal Transport System Optical fiber to central data processor Radius < 200 km 80 Gbit s -1 Data rate per antenna * Optimized layout, buried fiber, highly multiplexed. Radius > 200 km 80 Gbit s -1 Data rate per station * Leased and locally buried fiber connections. Central Data Processing System Correlator Input data streams 13, ants x 6 streams + 40 stns x 6 streams Frequency channels 10 5 Distributed as necessary over streams Complex Correlations 1.1 x ( /2)baselines x 4 pol n prod s x 10 5 chans Dump Time 200 ms Average dump time Beamformer Bandwidth 2 GHz Equivalent to one 2 GHz sub-band. Core-Inner beamformer 10 beamformers 1500 inputs, dual polarization, steerable on sky Time-domain Processor Pulsar / Transient Detector 20 One per output of beamformer system. 9

13 Science Computing Facility Input data rate 44 x Byte s -1 av ge from correlator (4-Byte real no s) Imaging Processor flops / input number (EVLA Memo 24) Archive 0.1 to 1 ExaByte Central Software Development 1000 PY Person Years Regional Science S/W Development 500 PY Person Years * The cost of lighting all the fibers in the data transmission system permits only 4 x 2 GHz sub-bands (each polarization). The Antennas and Correlator are sized for 6 sub-bands. Upgrading the data transmission system to 6 sub-bands would occur as soon as possible after construction. Table 2: Key System Performance Factors Antenna/Feed Efficiency 60% Coupled with feed/lna subsystems Minimum Elevation Angle 15 deg Set by ant. design; determines max. core FF Average T sys in low band ~40 K GHz, higher at low freq. end. Average T sys in high band ~35 K GHz, higher at high freq. end. System Equiv. Flux Density (each 911 Jy 60% efficiency, T sys = 35 K antenna) System Equiv. Flud Density ( Jy 60% efficiency, T sys = 35 K antennas) S (1") for 4 GHz bandwidth 3.4 µjy s -1/2 Min. detectable cont. flux (Stokes I), 3000 ant. 30 cm wavelength 1.37 deg 2 Scales as # 2 A e /T sys for T sys = 35 K m 2 /K Core (<0.5 km radius) ant.; Filling factor (FF) = TBD Core+Inner (<2.5 km radius) ant.; FF = TBD Core+Inner+Mid (<200 km radius) ant. All (<3000 km radius) ant. SSFoM ((A e /T sys ) 2 $), T sys = 35 K m 4 K -2 deg 2 Core+Inner+Mid (<200 km radius) 6.5 x ant. Imaging Dynamic Range ~73 db At frequencies less than 1.5 GHz Spectral Dynamic Range ~62 db 10

14 Offset-Gregorian optics (Figure 2) have been chosen for the SKA for the following reasons: 1. There will be no blockage of the main reflector and scattering of ground noise onto the main reflector can be minimized by enshrouding the feed/subreflector area. 2. Sidelobe levels will be significantly lower than for an axisymmetric antenna, which will increase the achievable image dynamic range over wide instantaneous fields of view. 3. There will be sufficient space for a feed changer and mechanical rotator for the phased array feed, if required. 4. Cross-talk between adjacent antennas is expected to be substantially less for an offset-fed antenna, since the feeds can be substantially screened, and the sidelobes and scattering will be significantly lower. Cryogenic dewars, cooled by Stirling cycle refrigerators will house the low noise amplifiers (LNAs), one per polarization for each feed, at the front-end of the receiver subsystems. The high band feed (1 10 GHz) will also be cooled. These refrigerators have been used with great success in the mobile phone industry, and have demonstrated excellent reliability. Cooling the high band feed, LNAs and coupling transmission lines will minimize the system noise temperature, and hence maximize the SKA s sensitivity and survey speed. Figure 2: A cross-section of the SKA antennas, showing the offset Gregorian optics, the location of the feeds (yellow), the mount, and a tracing of the ray-paths at the edge of the main reflector. This is a preliminary design and is still being optimized. The remainder of the receiver subsystem will be temperature stabilized slightly above the maximum ambient to maximize amplitude and phase stability. Outputs from the low band LNAs will be frequency converted to a quasibaseband for 4-bit digitization and optical modulation. High band outputs will be split into three 2 GHz wide sub-bands per polarization, to give a total instantaneous bandwidth of 6 GHz. Each of the sub-bands will be processed similarly to the low band channels, resulting in six 4-bit sample stream outputs per antenna. Use of digitized fiberoptic signal transport will ensure the best possible electromagnetic compatibility between subsystems, minimizing the potential for selfgenerated radio interference and maximizing immunity to external interfering signals. The digitized signals will also be unaffected by environmental and mechanical disturbance to the optical fibers. In addition to the dish-array, an aperture array will be considered as a potential transient detector, to be installed at a later date when the technology has reached a sufficient level of maturity. 11

15 Array Configuration Table 1 includes the fractional distribution of antennas in the array as a function of distance from the core-array center. The precise two-dimensional configuration of antennas is still being worked out and will of course be affected by terrain. The array configuration is closely related to the instrument transfer function of the telescope, and determines the range of spatial frequencies to which the telescope is sensitive. The array will not be reconfigurable. A large fraction of the proposed survey programs and other observations require distances between pairs of antennas (baselines) to be less than 200 km, and a few require much longer baselines. Except for a strong central condensation of baselines providing a high filling factor (the ratio of antenna collecting area to the area of the array within a given radius), the distribution of baselines will be approximately scale-free (self-similar) in two dimensions in the u-v plane. Considerable effort is being placed on positioning the antennas so as to produce a smooth 2-dimensional distribution of baselines, with as few gaps as possible in the u-v plane. Fortunately both prospective sites present few obstacles to achieving this goal in the 200 km diameter area. The strong central concentration of antennas is designed for pulsar observations in which the signals from the antennas will be summed into pulsar search and timing beams, and as well as to observe very low-brightness objects in imaging mode (see the section on the data processing system). For very high resolution observations, 25% of the antennas will be spaced from 200 km to distances of ~3000 km, close to the maximum possible distances in the Southern Africa or Australian continents (see Outer in Table 1). They may be located on 3-5 spiral arms emanating from the center, or they may simply be located in a two-dimensional array nearest to available fiber connections. The precise location of antennas at these distances is not critical. These antennas will be grouped into stations of about 20 antennas each, spaced a few antenna diameters apart. The signals from these antennas will be summed to form station beams, the outputs of which will go to the CDP via a fiber link. Forming beams in this way requires about 20 times less data to be transmitted to the center, but restricts the field-of-view at very high spatial resolution and marginally reduces the uv coverage. This will not greatly affect the science program because typically only one bright, sub-arcsecond object is expected to be found within the field-of-view of a single antenna. Signal Transport and Time Synchronization Signal Transport: Fiber connections from the antennas to the CDP within the 200 km radius will probably be buried, but may be suspended on overhead lines in some cases. The network will optimize total length, constrained by local obstacles. Fiber cables and power cables will share trenches wherever possible. The cost of burying the fiber dominates the total cost of data transmission. Therefore it makes sense to bury enough fiber for the foreseeable future, even though not all will be used initially. Data from the outer stations (beyond 200 km) will be sent on locally buried fiber from each station to a node on existing long-haul data networks and then sent on leased fiber along established routes to the central data processor. All of the equipment needed for data transport is available commercially, although specialized packaging will be needed to reduce the cost as much as possible. 12

16 Time Synchronization: Local Oscillators (LOs) and sample clocks for analog-to-digital converters (A/Ds) are required for the SKA RF systems contained in each antenna. These will be derived from a frequency synthesizer using an RF reference signal and timing signals to remove integer cycle ambiguities. The goal is to ensure that coherence loss on sub-second timescales (less than one correlator integration period) is limited to a few percent, and to limit the rate of change of phase so that large phase differences do not occur between calibrations (10 s of minutes). This requires the delivery of a reference signal with rms phase noise in the picosecond (ps) range and phase slope of <0.2 ps/min for 10 GHz operation. Methods of delivering these signals to antenna arrays using a round-trip feedback system have been developed for decades, and recently adapted to delivery via standard optical fiber systems for both the EVLA and for emerlin. Their requirements are comparable to those for the SKA for antennas at <200 km from the center. Refining these systems for the very large numbers of antennas will be carried out during the Preparatory Phase and continued during the Detailed Design Phase (as defined in the schedule in section V). Longer baselines may require more elaborate feedback systems or more stable oscillators at the remote ends. Optimized design and packaging will be needed for cost reasons, including the frequency synthesizers needed at the antennas. These are described in the RF package described under antennas. Central Data Processing System The Central Data Processing System (CDP) takes in ~14000 sampled data streams, at 20 Gbit/s each. Each of these must be divided into frequency channels and cross-correlated in antennabased pairs. Two types of processing will be carried out in real time: correlation and timedomain processing. Correlation is mature technology in radio astronomy. However, the SKA is a very large system, approximately 3800 times larger than the EVLA correlator, based on the product of the number of baselines and the total instantaneous bandwidth. The size does not imply a limit to feasibility, but its cost is likely to be a limiting factor. The maximum bandwidth that the SKA will ever need is 8 GHz per polarization (16 GHz total). The baseline design (Table 1) states 4 GHz per polarization, a reduction from 8 GHz to limit the size of the correlator. The science plan benefits from as much bandwidth as affordable. However, the plan can be accomplished with as little as 2 GHz per polarization. This could be viewed as contingency in the design, and will be explored in more detail in the section on cost. Development and Deployment of the CDP: An issue that historically has been a concern is the time needed to design and build and commission large correlators. This is because they are custom made systems, although they typically contain many repeated circuited boards and other components. Advances in digital technology have made it possible to tailor off-the-shelf technology for correlation in situations where the size of the system is sufficiently small. For example, this has led to the development of software correlators, running on generic computer clusters or small supercomputers, or on machines that utilize standardized circuit boards each containing a small collection of Field Programmable Gate Arrays (FPGAs). However, the SKA correlator will be sufficiently large that this approach would result in much higher cost and larger power dissipation than necessary. 13

17 The SKA correlator plan is staged to avoid the development time problem as much as possible, while at the same time providing a correlator system for all stages of the project. The first step of the plan begins in the SKA detailed design phase, when the correlator design team will begin the development of an architecture that can be executed using technology available for Phase 1, but upgradeable to effectively utilize technology available several years later. This approach will carefully select high-level designs of data paths, infrastructure (e.g. power supply, housing, cooling) and other major aspects that will be quasi-invariant as the system is upgraded, for example certain interface definitions. It will also require high-level, formal designs of processing units at the Register Transfer Language (RTL) level, using one of the high-level languages (e.g. VHDL). These designs will later be synthesized and executed in the appropriate hardware at execution time (e.g. FPGAs or ASICs). The team will also produce high-level designs of circuit boards and communication systems internal to the correlator. In parallel, a portion of the team will begin construction or acquisition of a test correlator, in time to be deployed along with the first few antennas. This will be a crucial piece of test equipment for the antennas. It could be a software correlator or a copy of one of the precursor correlators, and need not be able to process more than one 2 GHz band. The balance of the team will begin the execution of a Phase 1 Correlator, approximately 10% of the size of the final Phase 2 correlator. The technology-dependent detailed design will be carried out with the assistance of an industrial partner. Installation will be timed for deployment when a large fraction of the Phase 1 antennas are available. Although the Phase 1 correlator will be a fraction of the cost of the final correlator, much of the design cost will be allocated to both. As the Phase 1 correlator is being rolled out, detailed design of the Phase 2 correlator will begin, again with the assistance of an industrial partner. This will utilize a later generation of technology, but the same overall architecture and most of the high-level design as the Phase 1 correlator. The infrastructure will be designed so that the bulk of the Phase 2 correlator can be installed while the Phase 1 correlator continues to operate. Science operations will be continuous, since the Phase 1 correlator can be switched off as soon as the number of new antennas exceeds the number of Phase 1 antennas. A similar development approach will be used for the Beam-former and Time-Domain Processor. CDP Block Diagram and Description: Figure 3 is a block diagram of the CDP. Although the process is well enough understood to be able to derive 1 st order costs, this diagram itself is not final. It shows the key components of the real-time system, which are described briefly below. The architecture of the CDP is essentially that of an FX correlator, additionally incorporating beamforming and pulsar/transient detection. In an FX design the broadband input signals are first channelized into narrow frequency channels (the so-called F, or Fourier section), and signals in these channels are organized to flow to the X section (the Correlator), where signals from each antenna are multiplied (sample-by-sample) against those from all the other antennas. The result is time-averaged, typically for about 1 s before being dumped to an output buffer. The result is a data flow consisting of ~N 2 x n chan numbers per dump, where N is the number of antennas and n chan is the number of channels. These data are the fundamental input to sthe imaging subsystem (see Science Computing Facility). 14

18 "#$%&' () SKA Central Data Processing System, showing digital signals arriving from the antennas on fibers, a Channelizer/Beamformer, a Correlator, and a Pulsar and Transient Processor. The output of this entire system goes via optical fiber to the Science Computing Facility, which may be several hundred km away. In addition, the CDP incorporates a beamforming subsystem which sums the channelized signals from a selectable subset of the antennas. This has the effect of forming beams on the sky similar to what would be formed by a single antenna similar in size to the collecting area of the subarray of antennas. Many beams on the sky can be formed in this way from the same set of antennas. These beams can be searched for a variety of time-variable signals, using the pulsar/transient processor. In particular, this part of the CDP will be used as a hardware search engine for pulsars, a process that is usually carried out using general-purpose computers. For the SKA however, the high data rate will require hardware implementation of these search algorithms. Science Computing Facility Computing: The SCF will be physically located in a large population center, several hundred km from the SKA site. This is required to minimize the number of on-site staff, the cost of on-site infrastructure, the cost of on-site power and to keep a large source of RFI away from the telescope. It will be connected to the site by standard high-bandwidth optical fiber. Since the heaviest real-time signal processing will be done by the CDP, the SCF can be implemented using commodity high-performance computing systems augmented as appropriate with heterogeneous processing elements to achieve capital or power consumption savings, and by using generalpurpose software based on community or commercial components wherever possible. The requirement to image the entire field-of-view of the antennas determines both the maximum integration time (dump time) and the channel bandwidth for the correlator (see Table 1). Spectral line observations impose additional restrictions on the channel bandwidth. Note that although neither of these values affects the size of the correlator substantially, they have a direct impact on the size of the imaging system. The number of operations required to produce an image is approximately proportional to the number of u-v points (i.e. data rate) from the 15

19 correlator. Since the algorithms are iterative, the precise number of operations per data point is not deterministic, and must be measured. Measurements using the VLA indicate that floating point operations (flops) per point are needed. Using 10 4 flops per point indicates a requirement for a computer capable of 110 Petaflops/s (Table 1). Using the higher figure (10 4 ) provides some performance margin, but also takes into account the required input/output data rates, the amount of calibration processing needed and the very high imaging dynamic range required. Taking into account typical supercomputer utilization factors, defined primarily by parallelization efficiency, an Exascale supercomputer is likely to be needed (see also Q7 below) for the full SKA and a 10 Petaops machine for Phase 1. Archive: Historically radio telescope facilities have been able to archive raw visibility data. The SKA s 44 TeraByte/s (3800 PetaBytes/day) visibility data-rate precludes this possibility for two reasons: 1) even considering foreseeable advances in storage density, it is unlikely that any more than a few hours worth of visibility data can be buffered. 2) typical archival storage systems cannot sustain these read/write rates. Thus the archive will contain only data products images, sources, statistical information, etc. of the order of 100 PetaBytes to one ExaByte in volume. These will be archived at the SCF and replicated to the appropriate regional science centers globally. Large science teams will be needed to interpret these products and extract science results. These will be located at the regional science centers and in universities everywhere. Software Development: Seven areas of software engineering have been defined in the SKA project. 1) observer-focused (e.g. observation preparation) 2) telescope operations (e.g. scheduling) 3) system (monitor and control) 4) data handling, storage and distribution (including middleware) 5) calibration and imaging, 6) special data processing (e.g. for pulsar data) and 7) visualization software. These do not include engineering software, such as system software intimately tied to design or maintenance. Items 1 and 2 will be adapted from existing large telescopes and will consume only a small fraction of the total software effort. Item 3 is a substantial engineering effort, based on industrial standards, and will use both commercially available and existing telescope hardware and software subsystems wherever possible. Item 7 will be adapted from existing visualization software for astronomy and other application areas, taking advantage of hardware and software as they develop. Items 4-6 will require the largest, specialized software effort. Item 4 relates partly to archiving and distribution (see below), but also to ensuring that ancillary data are properly merged with the visibility data. Item 5 will require substantial new software development, with planning and requirements-development starting early in the SKA plan. Understanding SKA calibration and imaging on supercomputer systems is a specific focus of work being done by the U.S. TDP program. Item 6 will receive more planning attention during the next few years. The SKA will assemble a software team, starting in the Detailed Design Phase. It is estimated that the team will grow from ~20 people at the beginning of this phase to about 100 people. Based on studies of large industrial software projects, ~1000 PY of effort will be required to develop software for the SCF over 7-10 years. A major part of the SKA schedule will be surveys, with a smaller amount of time devoted to short follow-up or proposal-based observations. The central SKA organization will be mainly responsible for the handling, integrity, and distribution of data, as well as operating imaging pipelines and the like. Software development for the surveys will be allocated to the survey 16

20 teams, although this software may actually run at the SCF. Thus some of the people required for overall software development will reside at the regional science centers or wherever the survey teams are centered. Q2: Facility lifetime and upgrades: The SKA is designed as a general purpose radio telescope. The SKA facility with its environment protected from radio-frequency interference will continue as long as there is a need for cm-wave radio astronomy. It is expected that the actual telescope equipment and SCF computing and archive infrastructure will be upgraded regularly, especially the rapidly evolving electronics components. In the long term, even the antennas may eventually be replaced with other types of receptors, should they become available. Disposal will be in accord with current and future environmental regulations. Thus the site and its use for radio telescopes will have an indefinite lifetime. Q3: The general readiness and level of maturity of the technology: There are no technical barriers to detailed design/construction. Levels of performance may not be high enough to carry out all the planned science as rapidly as might be desired, and a few programs may not be fully realized. As noted in the answer to Q6, the project has recognized the areas where performance could be improved and project effort will be allocated accordingly. In the baseline design (excluding PAFs and AAs), the level of maturity and general readiness will be high after the SKA Preparatory Phase, when the PrepSKA-TDP work is complete. Q4 & Q8: Antenna construction methods: The 15-m diameter SKA antennas will be specified with rms surface accuracy (~ 1mm) and pointing specifications for at least 10 GHz operation, possibly 15 GHz. This is required to reach the Image Dynamic Range specification for ~1.5 GHz operation. There is no mass market now for such antennas. SKA and related R&D efforts have targeted mold-based fabrication techniques, which are well adapted to rapid, mass manufacture. Both metal and composite prototypes have been built and tested at 6-m and 15-m scales, respectively, which meet the rms surface accuracy requirements. Fabrication plants for either material can meet production rate targets. The metal reflector variety is molded in one piece and certainly requires a fabrication facility within a few hundred km of the telescope site. The composite reflectors may have similar requirements, although fabrication in several large pieces is likely to be feasible, which would enable fabrication anywhere. The rest of the antenna parts can be fabricated anywhere. Antennas will be designed for easy field assembly. Essentially no on-site adjustment of panels or other parts will be necessary and no factory pre-assembly will be needed either. Assembly of one antenna should be possible within a few hours. The SKA will build and exhaustively test one or more prototype antennas together with the EVLA, and develop a detailed design package ( build-to-print method) for the antennas. This will enable multi-sourcing of components without encountering a situation in which there are significantly different antenna-types in the array. Q5: Actuators and heritage: The SKA antennas will require azimuth and elevation bearings with standard actuators and bearings that are available from a variety of sources. The Imaging Dynamic Range specification requires pointing accuracy at 1.5 GHz to be about 0.01 beamwidth, similar to antennas made for 0.1 beamwidth pointing at 15 GHz (a typical specification for antennas). This specification is well within state-of-the-art for antennas and simply indicates that the SKA antennas must be relatively high quality in both surface accuracy and pointing. 17

21 Q6: Three components of lowest technical maturity: 1. WBSPF: Wideband single-pixel feeds, coupled with LNAs, have bandwidth ratios of at least 10:1, as compared with widely used feed/lna combinations with bandwidth ratios of no more than 2:1. WBSPFs on the antenna have T sys = 40 K and efficiency = 60%. More conventional 2:1 feeds can reach T sys = 28 K and efficiency = 70%. The U.S. Technology Development Program has been mounting an intense effort to improve the performance of WBSPFs (see Instruments section). 2. Software Development: Effective utilization of computing at extreme scales requires significant targeted investment in scalable, parallelized algorithms and their efficient mapping to underlying high-performance computing architectures, which are not static over time. Experience indicates clearly that this investment is needed if the transformational science enabled by extreme levels of computing performance is to be achieved in practice. Extensive research, development and planning is currently underway in these areas within the SKA technology development projects to address the relatively low technical maturity (see schedule). The risk mitigation strategies in these areas are well developed. 3. There are a few aspects of the telescope design at a similar level of technical maturity: 1) Details of reflector fabrication on a large scale. 2) Dynamic range capability of antenna systems. 3) The use of 20-nm chip technology in the CDP (only one step away from current technology). 4) Deployment of cryogenic cooling packages in large numbers. None of these aspects are risks at the technical feasibility level, but they may affect the cost or performance at a moderate level. All will be receiving considerable attention during the SKA Preparatory Phase (see schedule). Q7: Three greatest risks to cost, schedule and performance (apart from the international partnership): 1. Power Consumption Power generation in or transport to remote undeveloped sites is a large cost risk. Current estimates of on-site power consumption is expected to be at least 25 MW (~10 MW for the antenna array and ~12 MW for the central data processing system). Off-site power consumption for the SCF computing and archive infrastructure is not clear at this time, but could be >50 MW, in line with existing similar scale facilities. Innovative solar technology may be required, presenting a technical risk. The previous two factors could present a schedule risk, if delays in funding for power are encountered. The mitigation strategy is to establish a power audit committee to continuously assess power consumption during the detailed design phase. For example, passive thermal design is being investigated for antenna RF modules to obviate active cooling methods. Some performance trimming may be necessary. There are strong market pressures on all industries associated with digital information, concentrating on power efficiency and innovative power generation, especially for large-scale data centers. The SKA will certainly benefit from this concentration of industrial effort. 2. Radio Frequency Interference (RFI) The SKA will be sited in a remote area to provide the best possible RFI environment. However, RFI coming from the assembly of spacecraft always present above the sites 18

22 could limit the ultimate sensitivity if unchecked. It should be possible to excise occasional very strong RFI from satellites, which are easily detectable in the data. The mitigation strategy is to closely monitor the development of the precursor arrays (ASKAP and MeerKAT), which will be located on the sites, and encourage them to carry out long integration tests. RFI mitigation steps might be taken after the problem has been characterized, if it exists at all. 3. Software and algorithm development for future (100 petaflop) supercomputers that will be needed for the SKA. Commodity leading-edge, general-purpose supercomputer peak performance is predicted to be 50 Petaflop in 2015 and 1 Exaflop by The availability of such systems is not in much doubt, given market demand and prior industry delivery consistent with Moore's Law. Architectural trends supercomputer vendors may adopt on this time-scale are broadly known (e.g. very high levels of concurrency and heterogeneous processing elements), but are expected to evolve as vendors optimize factors important at this scale, including power consumption, programmability, and operations cost. Current radio astronomy imaging algorithms are well adapted to serial von-neumann computer architectures, and are being re-adapted to highly parallel and heterogeneous supercomputer architectures, with particular emphasis on very high effective utilization of the peak compute power of system architectures on the SKA time-scale. In particular, the use of reconfigurable Field Programmable Gate Arrays (FPGAs) as array coprocessors or other heterogeneous processing elements in High Performance Reconfigurable Computing (HPRC) architectures to improve computer system computational performance and reduce power consumption and cost will likely require new approaches to the development of radio astronomy imaging software. The mitigation strategy is to tailor the SKA science programs for the processing capability available at any given time, to wait for the appropriate software and computers to become available, and to re-use extreme-scale computing knowledge from other disciplines where possible. Much science is possible in the first few years of operation, utilizing only a portion of the final software. But on-going SKA effort will be needed to develop software. This cost risk thus also incurs a schedule risk. Q9: Specific accommodation of the telescope: Design for the core array will have to encompass desert conditions, strong sunlight, high ambient temperatures, and large diurnal temperature variation. Q10: Non-U.S. participation: The SKA has been a global project from the start. U.S. participation has been extremely high in a project with global contributions to the R&D and early design phases. The global partners are expected to fund ~2/3 of the overall cost. II.2 Observation Strategy Q1: Descriptive Overview: All of the key programs outlined in the Science Overview have survey characteristics: 1) long periods of observation will be required 2) many objects will be observed (millions or more) 3) scientific conclusions will result either from statistical analysis or from data mining through large numbers of objects, spectral signatures or events. Some of the surveys will require coverage of the entire observable sky; others will require deep integrations on relatively small areas of sky, but almost all will cover larger areas of sky than the 19

23 instantaneous field-of-view (FoV), defined in the baseline design by the size of the main reflector (Section II, Table 1). The rate at which an area of sky much larger than the FoV can be surveyed down to a limiting flux density is the survey speed SS=(Ae/Tsys) 2, defined in Section I. Nevertheless, some time will be needed for individual short observations, partly as survey follow-up. Since it may be difficult to carry out most surveys while the telescope is under construction, much of the first few years of operations will be devoted to individual short programs. Although it is too early to describe integration and mapping timelines in detail for the SKA, the scheduling will take into account three important factors: 1) science observations will begin as soon as significant collecting area is available, most likely with a program that is less sensitive to precise calibration (e.g. pulsar discovery). 2) Surveys that take more than a year of integration time will take at least two years of elapsed time. 3) Opportunities for carrying out more than one survey concurrently. For example, pulsar searching could be done concurrently with any other survey that covers the Galactic Plane. Ultimately surveys will have to be ranked on the basis of scientific importance and technical readiness. An estimate of the total integration time needed to carry out a representative list of surveys needed for the key science is about 15 years. Some of the surveys will need to be conducted frequencies less than 1 GHz, for which ionospheric conditions could affect the data quality or slow down calibration. Q2: Maintenance and Telescope Health: The SKA will be fault-tolerant: there are only a few single-point failure modes. Susceptible subsystems, such as central Local Oscillator distribution and certain control units, will contain redundant designs to permit maintenance and to reduce system down time. The exception would be a rare total power failure for which redundancy is not possible. A small fraction of the antennas can be out of service without significantly affecting performance, as can parts of the correlator subsystems, signal transport system, and the science computing facility. SKA antennas do not require the same 24/7 reliability as do Deep Space Network antennas, for example. The SKA will be equipped with a sophisticated monitoring system to track exception conditions and permit staff to repair out-of-service components. As part of the detailed design, a reliability analysis of each major component will be carried out to assess future operational service requirements and to enable design steps to be taken to adjust mean time between failures. This will enable capital and operational funds to be balanced so as to minimize total cost of ownership. Q3: The types of software and software development: have been described in the Array Configuration part of Section II. Q4: Diagrams: The most significant aspect of the antenna array is its configuration (see Section II/Array Configuration), the layout of antennas on the ground. It will be a centrally condensed array with stations of antennas beyond 100 km. A detailed configuration study will be carried out by the end of Although the emphasis in this document is on the dish array, the configuration will allow for the inclusion of a low frequency array and a dense aperture array in a multi-core configuration. Q5: Pointing: The SKA antennas must be able to point in any direction above an elevation limit of about 15 degrees. Pointing accuracy must be ~0.01 beamwidth at 1.4 GHz or ~0.1 beamwidth at 15 GHz, while tracking an astronomical position. This accuracy requirement can be met by 20

24 current technology; greater accuracy is possible by enhanced mechanical design. Fast slew rates will not be required, but observing time is wasted if the slew rates are too slow. The antennas described in Section II are Altitude-Azimuth style of antennas, the simplest and least expensive set-up for all sky coverage. Alt-Az antennas have a zone-of-avoidance at the Zenith, but this is not important. The antenna subsystem will contain an electronic motor control system that eliminates backlash and presents a standard interface to the rest of the system. Power-efficient designs of this type are widely available and no new technology is required. II.3 Facility Construction Q1: Siting: After a rigorous process, two sites, in the Karoo region of central South Africa and Murchison area in the state of Western Australia, were identified as excellent for the central region of the SKA. Radio quietness was a prime selection criterion for site short-listing in Both candidate central sites are very remote Southern Hemisphere desert locations with low population densities, now and for the foreseeable future. Both are situated in areas of high ionospheric stability, and have climatic water vapor characteristics supporting radio interferometer operation to at least 10 GHz. The core site in South Africa is an area with isolated mesa-like hills on one side and open, flat terrain on the other side, at an elevation ~1050 m high. The terrain in Western Australia is mostly flat, elevation ~450 m. Experience exists in development of such locations, and the site countries have already started the infrastructure development of the sites for small-scale precursor telescopes. Apart from power, the main infrastructure components needed to support the array are: antenna pedestals, access roads, fiber and power supply. The data processing building just outside the core, to which all fiber cables must be run, will be large enough to contain the correlator hardware, including routings and connections of incoming fibers and additional computing and data storage, allow efficient hardware cooling, and prevent the generation of radio interference. The active floor space required for this system will be m 2. Additional services inside the building will be required: office space, an array operations room, technical support space, and limited storage. This will require cooling systems, probably including liquid chillers, and mains power handling. On-site staff will be limited as much as is operationally feasible. The location of the Science Computing Faciity, away from the site, is the prime location for housing the majority of the technical staff: for Australia near the town of Geraldton, and for South Africa near Carnarvon. More detail on infrastructure is provided in the Basis of Estimate section under Infrastructure in the costing section. Q2: Electrical requirements: The SKA electrical power requirements have been presented as a risk item in Section II, Q7. Precise estimates for on-site power consumption are not yet available because minimizing power consumption of electronic equipment is an aspect of the detailed design. The most important on-site electrical loads are given by Table 3. A smaller amount of power will be needed for data transmission. System-level analysis and minimum-power designs will be used. As examples: antenna motors will be equipped with soft-start capability to minimize instantaneous loads, and antenna slew-starts will be staggered within the array for the same reason (tracking antennas present very small loads). Active cooling of antenna-based RF and electrical components will be avoided wherever possible, and phase-change materials may be used to shave peak temperatures within enclosures. 21

25 The project recognizes that low-power design will be a priority, and may even be a limiting factor in performance. A major reason for phasing the design and roll-out of the correlator is to maximize digital operations (e.g.flops) per watt, as well as operations per capital dollar. Both sites are desert areas with large fractions of clear days. Solar power-generation will be used wherever possible, particularly in situations where small amounts of power will be needed in remote locations (e.g. optical amplifiers), and may even be possible for individual antennas located in the km distances from the center. Solar generation is also being considered for supplying a much larger fraction of the total power. Power upgrades will have to be part of a general upgrade plan for the SKA in the future. The telescope itself is a separate power situation from the science computing facility (SCF), which will be located in a population center already served with power infrastructure. Load Table 3 Cooling Total from Supply Units MW MW MW MW Comments Infrastructure Misc. Services Antennas Antennas (passive cooling) Signal Transport Repeaters & terminal equipment Signal Processing Cooling for standard computing environment (TBC) Total On-site SKA Power 31 Q3: Specific infrastructure: Antennas will require foundations, whose cost has been allocated to civil works. The type of foundation will depend on local soil/rock characteristics, which will be determined by a geotechnical survey, to be carried out for both sites in the near future. In some locations antennas may be bolted directly to bedrock, while in others foundations of varying design will be needed. In all cases the foundation design will be limited by overturning moment in worst-case design winds (160 km/hr). Q4: Facility construction plan: Information on the facility construction plan is difficult to provide in detail at this stage of the project. For a general overview of the roll-out, see Section V (Programmatics and Schedule). Q5: New construction methods: The main non-standard construction and assembly aspect of the SKA is the minimization of radio frequency interference (RFI) on the site. This will mean that the buildings themselves will have to be built as partial shields and additionally contain shielded enclosures for such things as digital equipment. Construction techniques for incorporating metal cladding, grounding and shielded windows that provide some overall protection have been developed for new construction at a radio observatory in Canada. On-site assembly, equipment and other activities will be designed to limit RFI and require monitoring/inspection. Also, the production of reflectors for antennas will require near-site fabrication facilities. Two main mold-based fabrication options are being investigated which can produce either singlepiece reflectors or reflectors will a small number of assembly pieces. It is anticipated that 22

26 reflector production facilities will be located in the nearest population centers. Both sites have clear access for the transportation of very large objects from these centers. Q6: Construction management plan: The SKA has not yet reached a stage where the types of contracts or the style of management can be defined. This is being defined in the Preparatory Phase and fleshed out during the detailed design stage, when staffing of administrative and management positions will begin. 23

27 "#$$%&'()*+%','-.%$ "#"/$0)1&2&'+*3%&'()*+%'$/4$56'-*-7+8$9+:;+<'.($=%'+%%,&$ Q1: The baseline choice for SKA antennas is a clear aperture, offset paraboloid with Gregorian optics. Stable primary beam patterns (point spread functions) required for deep images with high dynamic ranges along with minimization of both spillover noise and susceptibility to radio frequency interference are the primary reasons for making this choice. The diameter (nominally 15m) is determined by several considerations: (1) minimization of total system cost of a large-n array taking into account antenna cost vs. diameter as well as electronics and digital processing costs; (2) consideration of candidate manufacturing methods for antennas to achieve economy of scale; and (3) a large-enough field of view to provide adequate survey speed for many of the science goals. The large-n array design itself provides a great deal of flexibility for deploying the collecting area. In particular, solid-angle coverage can be traded against sensitivity for very wide field sampling of the sky (100s of deg 2 ) for transient surveys. Q2: The reflector choice is based on ongoing extrapolation of demonstrated implementations. Off-axis reflectors are well understood. SKA antennas need to be optimized for particular feed antenna choices, which themselves are being designed and developed as described in the next subsystem section. Physical optics studies are now being done to identify optical designs that accommodate candidate feeds and optimize against system temperature, aperture efficiency, and cross polarization. Feeds and hence reflector optics will be selected in 2010 so that an SKA antenna outfitted with SKA feeds and receivers will be realized in Parallel development activity concerns the fabrication method for reflectors. Novel and improved fabrication techniques, design for mass manufacture and effective use of improved tooling and process control will all contribute to cost reduction. The most promising approaches involve composite materials and aluminum hydroforming, in both cases using a mold for monolithic dish fabrication. Composite fabrication methods are being pursued by our Canadian and South African partners as part of the current design effort (TDP, PrepSKA). An ongoing industry consultant study (Patriot Systems will provide to the project cost estimates for large-volume fabrication methods using different approaches. Information will also be provided from the ASKAP project, which is acquiring low-cost 12m antennas from a Chinese vendor (CETC54) that are a conventional on-axis design based on shaped metal segments. Achieving 0.5-1mm surface rms appears straight forward, although it needs to be demonstrated for 12-15m hydroformed monolithic surfaces. Development of a mount for any of the reflector options is based on extrapolation from a custom design for the Allen Telescope Array (6m reflectors). There is low risk in achieving the pointing requirement at the frequencies GHz) where it is most needed due to source confusion. The choice of a dual offset geometry is driven by optics requirements but also by a comparative cost study that suggests that the cost of a dual offset design is less than 1.5 times the cost of a prime focus antenna of the same aperture, for diameters of interest. The improved noise temperature performance alone compensates for much of this cost differential and the lower number of antennas reduces the computing cost of the array. All the proposed wide band feeds are low gain devices, they have wide angle illuminations. This leads to large subreflectors (roughly 3 meter diameter) which can be accommodated in an offset 24

28 design but would cause serious blockage and self shadowing in a symmetric design. Another benefit of the large subreflector is the good low frequency performance it provides. Fabrication: In the size and frequency ranges of the SKA antennas, structural rigidity and tolerance are not difficult. The required system RMS of about 0.5 mm. is achievable with standard practices and is routinely achieved in commercial antennas. Metrology techniques for measuring contours at this tolerance are readily available from commercial sources. The primary engineering challenge is to fabricate the reflectors in a cost effective manner. There are design studies under way or planned to evaluate composite reflectors, hydroformed reflectors and improved versions of present commercial practice. All of these methods hold promise of significant cost reductions. Feed Indexer: The proposed SKA antenna includes a feed indexer to interchange multiple feeds. This mechanism will add weight and cost to the antenna but is necessary for two main reasons. If a phased array feed is to be included as a future upgrade option then provision must be made to exchange between the PAF and one or more single pixel feeds. It would be essentially impossible to retrofit an indexer mechanism so it must be included initially. Also, it is very likely that more than one single pixel feed will be used to cover the SKA mid-band. Covering the entire 0.3 to 10 GHz bandwidth with one feed likely will not produce the desired performance. Subdividing the band gives better performance at the expense of more feed/receiver packages. Q3: The primary risks for reflectors are associated with manufacturing costs for the large number of antennas needed. Manpower costs depend strongly on the fabrication method and the country of origin. Materials costs are subject to large market fluctuations. Q5: Antenna assessment, design and development is a central work area of the U.S. Technology Development Project (TDP; ) that is part of the overall international design and costing effort coordinated by the SKA Program Development Office (SPDO; Manchester, UK). The TDP is led by Cornell University and involves a U.S.-wide effort with partner institutions Caltech/JPL, Minex, Inc., MIT/Haystack, Naval Research Laboratory, SETI Institute, UC Berkeley, and the University of Illinois. The Antennas Working Group of the TDP includes representatives from the SPDO and Australian, Canadian, and South African colleagues working on antenna design, fabrication and procurement. 25

29 Type of instrument Number of channels "#$%&'("$)*+,-(.)/(0-(1$2%#) Item Value Units Off-axis, dual-reflector antenna and azimuth-elevation mount Spectral Range (15 desired) GHz Number and Type of Sensors Number of Pixels Pixel size 3 feed systems 1 for WBSPFs, ~1000 for PAF Pixel scale deg Focal Plane Power and Thermal Requirements Temperature control range and accuracy Size/dimensions (for each instrument) Instrument mass without contingency (CBE*) 2000 (reflector, hub) Kg Instrument mass contingency % Instrument mass with contingency Kg (CBE+Reserve) Instrument average power without 2k per antenna W contingency (drives, cryocooler, analog elec) Instrument average power contingency % Instrument average power with contingency microns m x m x m Instrument average science data rate^ without - kbps contingency Instrument average science data^ rate - % contingency Instrument average science data^ rate with - kbps contingency Instrument Fields of View (if appropriate) - degrees Pointing requirements (knowledge) degrees Pointing requirements (control) 0.01 degrees Pointing requirements (stability) 0.01 deg/sec W 26

30 % "#"$%&'()*)+,-./)+0'-,/+%$1%234,(5/4%&3/67,893:,7%;,,4)%5/4%<,=,3>,0)% Q1: The wide frequency range ( GHz) for SKA-mid science needs to be covered with a minimum number of feeds and receivers owing to the large number of dishes that need to be populated. The current goal is to use wideband feeds that each cover up to a10:1 frequency range, as the two bands designated in Table 1, GHz and 1 10 GHz. For the lower band, the low-noise amplifier will be cooled with the feed at ambient temperature while the entire system will be cooled for the higher band using commercial off-the-shelf Stirling coolers. Three candidate feeds are under consideration in the TDP/PrepSKA design and costing effort and are at Technical Readiness Levels 6-7. These include the Allen Telescope Array (ATA) feed with an axial architecture, a commercial Lindgren Quad-ridge feed, and a Quasi-self-complementary (QSC) feed. Low-noise amplifiers (LNAs) now exist that can achieve low noise temperatures (<10K) over 10:1 bands when cooled to 20K. Current results with the ATA feed include system temperatures ~40-50K over a 7:1 frequency range, providing optimism for achieving system temperatures of 45K with the LNA at ambient temperature and 32K when cooled to 60K. Current work includes development of feed/lna combinations with good impedance match across the band and attention to integration details to lower the noise budget. The choice of feed will depend on performance specifications in 2010, including optimization with particular reflector optical designs, which depend strongly on the feed gain. Mechanical robustness will also be a selection criterion. Given current results and the existence of three candidate feeds (as well as a fourth being considered in Europe) we are optimistic that a suitable feed will exist for the SKA. Q2: Maturity levels are as follows. The LNAs are now essentially off-the-shelf items obtainable at low cost. Feed antennas range from being hardware demonstrated on other systems to extrapolation from demonstrated technology. Q3: The primary risk for feeds is achievement of low system temperature sufficient to satisfy science sensitivity and survey speed requirements. Current trends suggest that these risks will be retired by the end of If better performance is subsequently deemed important, a fallback position would involve use of three feeds instead of two, requiring a smaller total bandwidth per feed (e.g. 3:1) and anticipated better noise performance albeit at higher cost. Q5: WBSPF design and development is a central work area of the U.S. Technology Development Project (TDP; ) that is part of the overall international design and costing effort coordinated by the SKA Program Development Office (SPDO; Manchester, UK). The TDP is led by Cornell University and involves a U.S.-wide effort with partner institutions Caltech/JPL, Minex, Inc., MIT/Haystack, Naval Research Laboratory, SETI Institute, UC Berkeley, and the University of Illinois. Feed designs are monitored closely by the TDP Antennas Working Group. Work on WBSPFs is being done specifically at Caltech, Cornell, the SETI Institute, and Minex. 27

31 Figure 4: Example wideband feeds with log-periodic elements. From left to right are the QSC feed, the ATA feed, and the Lindgren Quad-ridge feed being developed or integrated respectively at Cornell, the SETI Institute (Allen Telescope Array), and Caltech. Figure 5: Measurements of a broadband LNA prototype. 28

32 "#$%&'("$)*+,-(.)/01(,+"1)20"3-(4506(-)7((1#) Type of instrument Number of channels Item Value Units Wideband Single Pixel Feed 2 polarizations Spectral Range and 1-10 GHz Number and Type of Sensors Number of Pixels Pixel size 2 wideband LNAs 1 per feed microns Pixel scale Determined by dish diameter arcsec Focal Plane Power and Thermal Requirements Temperature control range and accuracy Size/dimensions (for each instrument) Instrument mass without contingency Kg (CBE*) Instrument mass contingency % Instrument mass with contingency Kg (CBE+Reserve) Instrument average power without W contingency Instrument average power contingency % Instrument average power with contingency Instrument average science data rate^ without contingency Instrument average science data^ rate contingency Instrument average science data^ rate with contingency Instrument Fields of View (if appropriate) Pointing requirements (knowledge) Pointing requirements (control) Pointing requirements (stability) m x m x m W kbps % kbps degrees degrees degrees deg/sec 29

33 "#"$%&'()*+,&(%$-%.,&()/0%1/(/%2)34,''3)%5.126% An overview of the CDP is described in Section II, including a sketch of a development plan and an general outline of recent correlator development in radio astronomy. Details are described here on issues that touch on system design and feasibility. Figure 6 is a block diagram showing the key components of the CDP. For clarity, the figure focuses only on the main Correlation for the 2280 dishes. Although the process is well enough understood to be able to derive 1 st order thermal and financial costs, this diagram itself is not final. Figure 6: A schematic of the channelizer and correlator sections of the CDP, showing data-flow and processing requirements. The thermal dissipation figures are for active data processing and datatransmission components only, and do not account for the power dissipated by support chips, power supplies, and other ancillary components. Channelizer and Beamformer: Analog to Digital Conversion is implemented within each individual antenna for 2 polarizations and three individual sub-bands of 2 GHz each. Digitization is assumed to be 4 bits with a sample rate of 5 GHz, slightly oversampled from the Nyquist criterion by a factor of 2.5. Digitized data are passed through the Signal Transport network and arrive at the Channelizer and Beamformer combination as detailed in Figure 1. For this implementation it is assumed that 100 Gbit/s links will be a viable option. This is a likely scenario as the IEEE 802.3ba specification is due to be ratified in Overall 4560 fibers, each carrying up to 100 Gbit/s data links, are received from the dishes with an extra 80 links from the Stations. Once time-aligned, the antenna data streams are split into frequency channels using a two stage poly-phase filter, which produces a complete set of frequency channels for correlation and/or pulsar/transient processing. 30

34 Frequency-domain beamforming is performed across the 600 dishes in the 1-km diameter core to produce beams for use within the pulsar processing and long baseline correlation. This is implemented as a complex coefficient multiply and sum across the antennas. Communication Fabric: A communications fabric is utilized to distribute the data streams from the Channelizer/Beamformer to the individual processors within the Correlator and Pulsar/Transient processors. This utilizes a passive optical interconnect unit for signal routing. For the main part of the Correlator this consists of 72 x 38 point-to-point fibers. Additional links are required for the Long Baseline Correlator and Pulsar processing. A switched Gigabit Ethernet fabric is utilized for Monitoring and Control (available commercially). Correlator: The cross-correlation provides all Stokes products, across all baselines for 2280 dishes. Additionally, long baselines are correlated between stations and beams generated from the central core. Cross-correlation products are integrated over a period whose maximum is determined by two factors: Fringe Nyquist rate on the edge of the processed field-of-view, and limiting reduction in the visibility amplitude due to time averaging to less than 10% for the longest baselines. Of these, the latter smearing has the greatest impact and limits the integration period to less than 200ms, giving a total dump rate of 2 x 1013 thirty-two bit values per second. Table 4: Processing Efficiency Projected for 2012 Technology Giga Operations per Watt x86 (3.2 GHz Octocore) 0.2 Graphics Processing Unit (NVidia Tesla) * 5 Field Programmable Gate Array,FPGA 25 Application-Specific Integrated Circuit, ASIC 400 * Available in Thermal Dissipation: Thermal dissipation in a system of this size is a key design criterion because it can dominate both the capital and operating costs of the system. Table 4 outlines the processing power per Watt expected in 2012, projected from manufacturer s published performance figures. These dissipation figures account for both dynamic and static dissipation contributions. Dynamic dissipation improves with silicon feature size whereas static dissipation gets worse due to current leakage. This is the case for all the processing technologies. In this system, the correlator carries the largest processing load, due to its dependence on the square of the number of antennas. The dissipation figures given in Figure 6 above have been calculated assuming an ASIC implementation that utilizes a carefully designed 4-bit multiplier, which will take less than 500 gates. As a rough approximation for noise signals, it has been assumed that less than half its transistors are switching at any one time. The data in Table 4 show that an ASIC-based solution has significant power-consumption advantage for a given amount of processing. ASIC manufacturers are already beginning to offer 30-nm lithography, and by 2012, 22-nm technology will be available at less than 2.5 nw/mhz/gate. We assume that within the timescales of the SKA, this technology will be suitably mature. Non-recurring engineering (NRE): NRE costs for ASIC implementation are the most significant of all of the NRE charges in the CDP. The NRE cost of such state-of-the-art devices can 31

35 normally only be justified by large production quantities. (It currently costs ~4M for 30nm lithography). Although the number of devices is not nearly as high as those expected for the mass market, the expected NRE of ~300 per die will be quickly offset by savings in the cost of power for operations (i.e. The factor of 16 improvement in power efficiency of ASICs over FPGAs would pay for the difference in ~2 years). Specific Questions: Q1: Describe proposed science instrumentation: see Section II (Technical Implementation) and the above description. Q2: Technical maturity, historical context, and development plan: See answers to Q6, Q7 in Technical Implementation. Q3: Technical Risks: Three signification risks: 1) 20-nm technology might not be available on the time-scale required. 2) Non-recurring engineering costs are higher than anticipated. 3) The development time for the correlator is longer than anticipated. Q4: Instrument table: The format of the instrument table does not fit the CDP, and so a diagram has been substituted. Q5 & Q10: Correlator development has a long history in radio astronomy. The most recent, large correlators were built for ALMA and the EVLA. These projects were carried out by groups with expertise in the large digital processing machines. As outlined in Section V (Progammatics and Schedule), the SKA organization will have a dedicated team for digital signal processing, in addition to using industry for assistance with production. Q6: The EVLA correlator project has a very extensive set of design and development documents, although they are currently not publically available. The National Radio Astronomy Observatory in the U.S. and the National Research Council of Canada, the organizations that supported these developments, are members of the SKA global consortium. Q7: The CDP will have many operational modes. In the case of the correlator the modes are mainly related to the number of channels and the number of subarrays being processed. Calibration of the correlator is not required. The number of bits and the data volumes are outlined above. Q8: There will be an instrument control component to the correlator project. There is considerable experience in correlator control developed for previous correlator projects. Q9: The SKA is an international project, which inherently requires both U.S. and non-u.s. participation. 32

36 "#"#$$%&'()(*+,-.(*/&,+.*$#0$123(+456//3)$7++4$81679$ Q1. A PAF is a possible upgrade instrument for mid-frequency SKA reflectors to increase both the field of view (FoV) and aperture efficiency and thus increase survey speed beyond that provided by instrument 2, the wideband single pixel feed (WBSPF). The target frequency range is GHz to address key science in dark energy, cosmic magnetism, and radio transients. Incorporation of PAFs is subject to their meeting bandwidth and T sys performance specs. The PAF is a dense array of elements, separated by /2 for the longest wavelength, each with an LNA, frequency converter and signal digitizer. Digital beam forming corrects for aberrations that improve the off-axis aperture efficiency by up to 40% and provides multiple pixels that can increase the solid-angle coverage by a factor ~30. Each beam produced by the PAF requires signal transport to a correlator where corresponding beams of different antennas are correlated. A primary application of PAFs is fast surveys. Uncooled PAFs are projected to yield T sys ~ 50K compared to 30K for the WBSPFs, increasing the survey speed by a factor ~20. Cryogenic cooling of PAFs is being considered and would further enhance survey speed but cooling the entire feed would entail a very large and costly cryostat. PAF costs are dominated by digital signal processing and signal transport, which continue to decrease a factor ~2 every 1.5 to 2 years due to commercial development of these technologies. Further cost reductions will ensue from use of custom ASICs for the receiver and beamformer. Q2. Since interferometer dish arrays using PAFs do not exist, the PAF instrument is a new technology for radio astronomy. Prototype PAFs are currently being developed by ASTRON, BYU/NRAO, DRAO and CSIRO. These have been constructed mainly from off-the-shelf components but have required very substantial and ongoing efforts in analysis and design in order to approach the desired performance. The Australian SKA Pathfinder (ASKAP) is developing a 36-antenna PAF interferometer and the WSRT in the Netherlands is being equipped with PAFs as well. These will be testbeds for new calibration techniques and approaches to obtaining the required beam stability. Table 5 summarizes recent PAF test results. Further development clearly is required to reduce T sys to 50K. Good performance with uncooled systems is based on recent LNA results, summarized in Table 6, and theoretical studies of optimum noise matching of arrays and LNAs. The measured T sys of 66K for the narrowband BYU/NRAO PAF should be reduced to less than 50K by optimum matching, eliminating most of the 20K attributed to noise mismatch and inefficient coupling in the current system. CSIRO results provide an indication that this performance can be achieved over a 2:1 frequency range Table 5: Measured PAF System Temperatures and Aperture Efficiencies Organization Array LNA T min T sys (K) ap T sys / ap (K) (K) BYU/NRAO 19 element dipole 1.6GHz ASTRON 112 element Vivaldi, 1.4GHz CSIRO 40 element Checkerboard, 1.5GHz 40 NA NA

37 Table 6: Measured Room-temperature LNA Noise Temperatures Organization LNA T min (K) ASTRON Single-ended, 50ohm,1-1.8GHz 40 CSIRO Differential, 300ohm, GHz 40 BYU/NRAO Single-ended, 50ohm, 1.6GHz 33 U. Calgary Single-ended, 85ohm, GHz U. Calgary Differential, 100ohm, GHz 29 Q3. A technical and cost risk is achieving the target system noise temperature T sys ~50K and coupling efficiency between the LNA and the elements of the PAFs. Modeling suggests this can be achieved over a 2:1 frequency range. Cryogenic cooling could reduce T sys to 30K but at too high a cost due to the many distributed PAF elements and large size of the overall PAF aperture. A second technical risk involves beam calibration and stability sufficient to achieve the required dynamic range 10 6 :1 for spectroscopic surveys. Optimally, the beams need to be fixed and stable with respect to sources on the sky, requiring either mechanical rotation of the feed or entire reflector, or electrical rotation of the PAF beams by appropriately programming the beamformer weights. Whether electronic rotation can be achieved with sufficient precision is yet to be determined. ASKAP will test electronic rotation against a mechanical solution. Q4. See attached instrument table. Q5. The knowledge gained in PAF development activities worldwide is being shared with the international SKA community. Q6. See for a description of ASKAP. Q7. The PAF requires both survey and calibration modes. For surveys, PAF beams track and remain constant with respect to sources in a given region of the sky for deep integrations exceeding100hrs duration. Calibration techniques are being developed by the international SKA community (see Q9). One approach is to modulate beamformer weights with orthogonal sequences that allow each PAF element to be correlated against a phased array formed by all other antennas in the interferometer, thereby maximizing the signal-to-noise ratio of the calibration measurements and minimizing the calibration time. Similar techniques similar have received some study for the Deep Space Network but further work is required for the SKA. Q8. Control software is required to modify weights in real time, especially if real-time beam rotation is required. In addition, the software will monitor system behavior, switch between survey and calibration modes and implement the calibration algorithm parameters. At this early development stage, a full suite of software is not yet available. Q9. SKA implementation of the PAF requires further development and testing in interferometer contexts. The U.S., Netherlands, Canada and Australia are playing leading roles in this work. All are members of the international SKA consortium. Q10. Many of the PAF components have been used in similar antenna arrays. For example, Vivaldi-element PAFs being developed at ASTRON and DRAO have been adapted from wideband arrays used in radar. Off-the-shelf LNAs are typically too noisy for this application at ambient temperatures. SKA partners are working on ambient-temperature LNAs with some success compared with off-the-shelf LNAs (see Table 6). Most of the other RF components are commercially available. Specialized digital hardware will have to be developed ultimately to permit the PAF processing to occur in a cost and power-efficient manner. 34

38 "#$%&'("$)*+,-()./%)"#$%&'("$)01)23+#(4)5%%+6)7((4)82579) Item Value Units Type of instrument Phased Array Number of channels channels 18.5KHz wide Spectral Range GHz Number and Type of Sensors 192 antennas 6-36 dual-polarizied digital outputs Number of Pixels 6-36 Pixel size NA Pixel scale 0.9 at 1.8GHz deg Focal Plane Power and Thermal Requirements 552W uncooled Temperature control range and accuracy C Size/dimensions (for each instrument) 1.1x1.1x1 mxmxm Instrument mass without contingency (CBE*) 200 Kg Instrument mass contingency 20 % Instrument mass with contingency (CBE+Reserve) 240 Kg Instrument average power without contingency Receiver: 552 (focus) W 720(pedestal) Beamformer: 1200(pedestal) 2000(station or correlator) Total: 4572 Instrument average power contingency 20 % Instrument average power with contingency 5.5 kw Instrument average science data^ rate without 612 Gbps contingency Instrument average science data^ rate contingency 20 % Instrument average science data^ rate with 735 Gbps contingency Instrument field of view (if appropriate) 30 degree 2 Pointing requirements (knowledge) 30 arcsec Pointing requirements (control) 30 arcsec Pointing requirements (stability) 30 arcsec *CBF = Current Best Estimate ^Instrument data rate defined as science data rate prior to processing. 35

39 "#"$%%&'()*)+,-./)+0'-,/+%$1%2,/),%34,0+'0,%3005*% Q1. In a dense aperture array (AA) collecting area is formed from a large number of electrically small, closely spaced antenna elements plus receiver chains in a fixed array. A dense AA is illustrated in Figure 7 with Vivaldi wideband feed elements, each of which is about a wavelength in height, spaced a ~1/2 wavelength from each other. Digital beam-formers sum signals after compensating for path-length delays to form a single output. Beam-sizes formed by AAs and dishes of the same diameters are approximately the same size. However, multiple beams can be formed simultaneously from all or a subset of elements in an AA using a beam-former. Aggregate fields-of-view of hundreds of deg 2 are potentially accessible. Although AAs have potential for many uses, the most scientifically interesting near-term use is to monitor large areas of sky for transients, with subsequent triggering of the dish-array and other telescopes. Beams formed within the ±45 cone of sensitivity from individual elements allow rapid scanning of large areas of sky. More beams means larger total FoV. A small AA is well suited to searching for bright radio transients, such as prompt radio emission from gamma bursts, flare stars, and giant pulses from pulsars. AA Specifications: With current technology AAs are best suited to frequencies of MHz. An AA a collecting area of ~10,000 m 2 is envisaged, operating over the above frequency range. T sys of 50 K (a little higher at 300 MHz) and an antenna efficiency of 80% is possible, depending on the outcome of system testing (see below). There is flexibility in dividing the array into independent patches. A frequency channelizer is needed so that transients can be located in frequency and that plasma dispersion can be removed. Using strong pulsars as a guide, at least 1000 frequency channels would be needed. The parameter search space for transients is wide open allowing gradual development of the AAtransient detector as time and resources permit. For example, fewer beams or a narrower bandwidth would reduce the initial cost. The selection of an upper frequency limit of less than 1000 MHz reduces the initial cost as the square of the selected upper-frequency ratio. Unfortunately it is difficult to improve T sys or the aperture efficiency once the array is built. These will be the most critical areas of development in the next few years. LNAs for AAs cannot be cooled, so low-noise, room temperature receivers are a basic requirement. The cost goal for the transient monitor is ~$2600/m 2, including the beamforming equipment but excluding the Pulsar/Transient processor. At this price the transient monitor will cost ~$39M. Detailed cost and power modeling will be delivered from the SKADS work in Q2,Q6,Q10. Technical maturity and heritage: Mid-frequency dense arrays, operating 300MHz to >1000MHz with 160 m 2 collecting area will be demonstrated by the end 2009, as the culmination of SKADS, a European SKA design study of AAs: see This work shows that dense AAs are becoming practical and that the electromagnetic performance of 36 Figure 7 Aperture array demonstrator, showing some of the many Vivaldi elements which make up the system.