SKA Square Kilometre Array

A telescope that spans two continents

The Square Kilometre Array is a single radio telescope split across two continents, on purpose. One half — SKA-Low — sits on the Murchison radio-quiet zone in Western Australia and consists of roughly 131,072 log-periodic dipole antennas (essentially fishbone-shaped wire antennas planted in the red dirt) arranged in 512 stations of 256 antennas each. It listens between 50 and 350 MHz. The other half — SKA-Mid — sits on South Africa's Karoo plateau and consists of 197 parabolic dishes: 133 newly built 15 m SKA dishes joined to the 64 existing MeerKAT dishes. It listens between 350 MHz and approximately 15 GHz. Both halves are correlated together as a single interferometric instrument, headquartered out of SKAO at Jodrell Bank in the UK and run by an intergovernmental treaty organisation with eight founding member states.

The name is a goal as much as a description. The 1990s concept paper specified one square kilometre of total collecting area as a round-number sensitivity target — about 50 times the best radio telescopes then in operation. The realised Phase 1 design comes close to that target across the bands: SKA-Low contributes about 0.4 km² of effective dipole area at 100 MHz, SKA-Mid roughly 35,000 m² of dish area times its filling factor and frequency-dependent efficiency. The combined survey speed and sensitivity is the loose sense in which "square kilometre" is accurate.

Why the same telescope needs two designs

The split between SKA-Low and SKA-Mid is forced by physics. A parabolic dish only works efficiently when the dish diameter is several times the observing wavelength: a 15 m dish at 1 GHz (λ = 30 cm) is a fine collector, but at 100 MHz (λ = 3 m) the dish is only a few wavelengths across — diffraction limits its beam to something useless. Conversely, an aperture array of phased dipole antennas works beautifully at long wavelengths but degrades at high frequencies, where the spacing between elements becomes comparable to the wavelength and grating-lobe contamination explodes.

So below ~350 MHz, SKA-Low uses an aperture array — thousands of fixed dipoles whose signals are combined in software to electronically steer a synthetic beam anywhere on the visible sky, instantly, with no moving parts. Above 350 MHz, SKA-Mid uses mechanically steered dishes in a classic radio-interferometer configuration. The crossover frequency around 350 MHz is the practical sweet spot.

Splitting the project across two host nations also has political and operational advantages. Australia and South Africa had the world's two largest legally protected radio-quiet reserves, both with substantial precursor facilities already on-site (the MWA and ASKAP in Murchison; MeerKAT in the Karoo). Folding those precursors into the larger SKA design — MeerKAT becomes part of SKA-Mid; MWA continues to operate alongside SKA-Low — leveraged a decade of prior investment.

How a radio interferometer works (very briefly)

A radio interferometer doesn't form an image directly. Instead, every pair of antennas in the array measures the cross-correlation of the incoming wavefront from a single point in the sky:

V(u, v) = ∫∫ I(l, m) exp(-2πi (ul + vm)) dl dm

V        complex visibility measured by one antenna pair
(u, v)   antenna pair separation projected onto the sky, in wavelengths
I(l, m)  sky brightness at direction cosines (l, m)

The visibility is the Fourier transform of the sky brightness, sampled at the spatial frequency (u, v) of that baseline. With N antennas you get N(N−1)/2 baselines — for SKA-Mid's 197 dishes, that is 19,306 simultaneous (u, v) samples — and Earth rotation over the course of an observation rotates each baseline through (u, v) space, filling in more samples. A computer then inverts the Fourier transform to recover I(l, m), the sky image. The angular resolution is roughly λ/B_max, where B_max is the longest baseline; for Phase 1 the longest SKA-Mid baseline is 65 km, giving sub-arcsecond resolution at GHz frequencies. The future Phase 2 extends baselines to thousands of kilometres.

The five science questions SKA was designed to answer

SKA is, unusually for big-physics projects, an open-purpose facility — it does whatever the community asks — but the construction case was built around five flagship science drivers, each picked because no other instrument could do it.

1. The epoch of reionization (SKA-Low)

The neutral hydrogen that filled the universe between recombination (z ≈ 1100) and the rise of the first ionising sources emitted a faint 1420 MHz (21 cm) spectral line. That line is cosmologically redshifted into SKA-Low's band: z = 6 corresponds to 200 MHz, z = 15 corresponds to 89 MHz, z = 27 to 50 MHz. By tuning across the band, SKA-Low builds a redshift-sliced tomographic map of how the early intergalactic medium got ionised — bubbles of ionised gas growing around the first stars, quasars, and dwarf galaxies and eventually overlapping into the fully ionised universe we live in. No other instrument can do this — the signal is buried under foregrounds 10⁴ to 10⁵ times brighter, and only the SKA's combination of sensitivity, baseline coverage, and per-channel calibration can dig it out.

2. Pulsar timing arrays and the nanohertz GW background (SKA-Mid)

Millisecond pulsars are spectacularly stable clocks — better than 100 ns over a decade in a few well-studied cases. A nanohertz-frequency gravitational wave passing between Earth and a pulsar slightly perturbs the intervening spacetime, shifting the pulse arrival time at Earth by tens of nanoseconds. The signature is a quadrupole correlation between pulsars at different sky angles (the Hellings-Downs curve). NANOGrav, EPTA, and PPTA together announced evidence for this stochastic background in 2023 using 15 years of timing on about 70 pulsars. SKA-Mid will time hundreds of pulsars at higher cadence and lower per-arrival-time noise, reaching ~100× the timing sensitivity and resolving individual supermassive-black-hole-binary sources rather than just the background hum.

3. Fast radio burst host galaxies (SKA-Mid)

Fast radio bursts are millisecond pulses of bright coherent radio emission. The community has gone from 1 burst (Lorimer 2007) to thousands (CHIME, ASKAP) without firmly settling on the origin — magnetars are the leading suspect after the 2020 burst from SGR 1935+2154, but there are clearly multiple populations. The bottleneck has been precise localisation: most detectors locate to an arcminute, not enough to pin a specific galaxy. SKA-Mid's sub-arcsecond resolution combined with millisecond voltage buffers means a burst can be localised to within its host galaxy in real time, then deeply imaged for a persistent radio counterpart, and cross-matched against optical surveys for environment.

4. Cosmic magnetism (both telescopes)

The origin of magnetic fields in galaxies and clusters is a 100-year-old mystery. SKA will measure the rotation measure — the Faraday rotation of polarised emission by intervening magnetised plasma — toward roughly 10 million extragalactic polarised sources, building a "rotation-measure grid" of the magnetic field of the whole observable universe. This breaks degeneracies in cosmological seed-field models and tests whether primordial fields from the early universe survive.

5. Cradle of life: protoplanetary disks (SKA-Mid)

SKA-Mid at high frequency (10–15 GHz) reaches sub-AU resolution on the nearest forming planetary systems. Combined with ALMA (which dominates at sub-millimetre), SKA closes the wavelength range over which the dust and gas in protoplanetary disks emit, and uniquely traces the centimetre-wavelength growth of pebbles into planetesimals — the bridge across "the metre-size barrier" between dust grains and planet-scale bodies that has frustrated planet-formation modelling for decades.

SKA-Low in detail

SKA-Low covers an arid plateau in the Murchison region of Western Australia, jointly with the existing Murchison Widefield Array and ASKAP. The array consists of 512 stations distributed over baselines up to 65 km. Each station is an irregular cluster of 256 log-periodic dual-polarisation dipole antennas covering 38 m in diameter. The total antenna count is 131,072.

SKA-Low parameter	Value
Frequency range	50 – 350 MHz
Number of antennas	~131,072 (256 antennas × 512 stations)
Station diameter	38 m
Maximum baseline	~65 km
Angular resolution at 200 MHz	~7 arcsec
Field of view at 100 MHz	~330 deg² (huge — solid-state beamforming)
Sensitivity (system equivalent flux density)	~1.7 Jy at 110 MHz

Each antenna's signal is digitised on-site, beam-formed at the station level into one or more synthetic beams, then sent by fibre to the Central Signal Processor in Perth. Because the antennas are fixed and the beam is steered electronically, SKA-Low can stare at multiple parts of the sky simultaneously — every station beam is independent. This is what makes the all-sky reionization survey possible: SKA-Low integrates on every accessible piece of sky in parallel.

SKA-Mid in detail

SKA-Mid sits on the Karoo plateau in South Africa's Northern Cape, the most radio-quiet large landmass in the Southern Hemisphere. It is technically two telescopes folded into one: 133 new SKA dishes (15 m diameter) plus the 64 existing MeerKAT dishes (13.5 m). The combined array of 197 dishes is correlated together at the Central Signal Processor in Cape Town.

SKA-Mid parameter	Value
Frequency range	350 MHz – ~15.4 GHz (Phase 1 bands)
Number of dishes	197 (133 new SKA + 64 MeerKAT)
Dish diameter	15 m (SKA) / 13.5 m (MeerKAT)
Maximum baseline	~150 km (Phase 1)
Angular resolution at 1.4 GHz	~0.3 arcsec
Field of view at 1.4 GHz	~1 deg² (single pointing)
Sensitivity (1.4 GHz)	~1 μJy/beam in 1 hour

Operationally, MeerKAT was already producing world-class science before SKA-Mid construction began — it discovered the Vela X-1 stellar wind, found radio bridges in galaxy clusters, and contributed key millisecond pulsar timing data. Folding it into SKA-Mid lets the larger array start observing immediately with a smaller dish count and grow to full sensitivity by 2030.

The data problem

SKA's raw data rate is roughly a terabyte per second per telescope — more than the entire global internet backbone traffic when the project was proposed. There is no plan to store this: the architecture deliberately pushes data reduction to the edge.

• On-station digitisation. Each antenna or dish digitises its voltage stream locally — at SKA-Low's antennas, dual polarisation × 16-bit × 800 MS/s × 256 antennas per station yields ~6 Tb/s per station. Each station beam-forms down to ~1.6 Tb/s before going on fibre.
• Central Signal Processor (CSP). The CSP at each telescope cross-correlates beams or dishes, producing a stream of complex visibilities at roughly 1 TB/s.
• Science Data Processor (SDP). The SDP at each telescope ingests the CSP output and performs calibration, imaging, source extraction, and time-domain triggering — including buffered voltage capture for FRB localisation. It compresses the stream by a factor of about 1000× into ~300 PB/year of science-ready data products.
• Regional Centres. Science-ready products are distributed to a federated network of SKA Regional Centres in member countries, where end users actually access them. Each Regional Centre is itself a multi-tens-of-petabyte facility.

The total on-site compute budget is on the order of 100 petaflops sustained per telescope — comparable to a small national supercomputer, but dedicated to a single instrument. The architecture is one of the few cases where the data-handling design is as ambitious as the radio-frequency design.

Timeline and partnership

The SKA story has been a 30-year diplomatic exercise as much as an engineering one. Concept papers in the 1990s; site selection in 2012 (split, after the South Africa vs Australia debate, to use both); detailed design 2012–2019; intergovernmental treaty organisation (SKAO) established 2019; construction approved 2021. Operations are headquartered at Jodrell Bank in the UK — symbolically the home of British radio astronomy and the site of the 76 m Lovell Telescope from 1957.

Phase	Year	What
SKAO treaty signed	2019	Intergovernmental organisation established at Jodrell Bank
Construction approved	June 2021	Council greenlight; ground-breaking December 2022
AA0.5 (4 SKA-Low stations, 4 SKA-Mid dishes)	2024	First on-sky correlator test
AA1 (16 stations, 8 dishes)	2025	Software pipelines validated
AA2 (64 stations, 64 dishes)	2026	Early science calls begin
AA3 (256 stations, 144 dishes)	2027–2028	Major science output starts
AA4 / full Phase 1 (512 stations, 197 dishes)	2028–2030	Design sensitivity reached
Operations baseline	2030+	50-year operational horizon

The eight founding members are Australia, South Africa, the United Kingdom, the Netherlands, Italy, Portugal, Switzerland, and China. Several more nations (Spain, Germany, Sweden, France, Canada, India, Japan, South Korea, among others) participate as observers or are in accession. Construction cost is roughly €2 billion in 2021 currency; lifecycle (50-year) cost is several times that.

Precursors and pathfinders

SKA was never going to be designed and built cold. Instead, the project explicitly funded "pathfinders" and "precursors" — smaller instruments that test SKA technology and answer pieces of the SKA science case. The two co-located precursors are particularly important: they will be folded into the SKA telescopes themselves.

• MeerKAT (South Africa, 64 dishes). Operating since 2018; folded into SKA-Mid. Its 13.5 m offset-Gregorian dish design is the template for the new SKA-Mid 15 m dishes.
• ASKAP (Australia, 36 dishes with phased-array feeds). Operating since 2012; tests survey-speed enhancement via phased-array feeds. Will continue to operate alongside SKA-Mid as a separate Australian SKA Pathfinder.
• MWA (Australia, 4096 dipoles). Operating since 2013; the original aperture-array precursor for SKA-Low. Continues to operate; SKA-Low builds on its calibration and imaging pipelines.
• LOFAR (Netherlands and Europe, 38+ stations). European low-frequency aperture array; tests SKA-Low technology and software at smaller scale.
• HERA (South Africa, 350 antennas). Hydrogen Epoch of Reionization Array — purpose-built 21 cm cosmology precursor.

How SKA compares to other observatories

Instrument	Band	Aperture / N elements	Max baseline	Best resolution	Status
SKA-Low	50–350 MHz	131,072 dipoles in 512 stations	65 km	~7" at 200 MHz	Under construction
SKA-Mid	0.35–15 GHz	197 × 15 m / 13.5 m dishes	150 km	~0.3" at 1.4 GHz	Under construction
JVLA	1–50 GHz	27 × 25 m dishes	36 km	~0.05" at 43 GHz	Operating
ngVLA (planned)	1.2–116 GHz	263 × 18 m dishes	~9000 km	~1 mas	Design, ~2035
ALMA	84–950 GHz	66 dishes (54×12 m + 12×7 m)	16 km	~5 mas at 950 GHz	Operating
MeerKAT	0.58–3.5 GHz	64 × 13.5 m	8 km	~5" at 1.4 GHz	Operating (→ SKA-Mid)
LOFAR	10–250 MHz	~50,000 dipoles in ~52 stations	1900 km (international)	~0.3" at 240 MHz	Operating
FAST	0.07–3 GHz	1 × 500 m fixed dish	n/a (single dish)	~3" at 1.4 GHz	Operating (China)

SKA-Mid's collecting area is comparable to JVLA's but with sevenfold more dishes and substantially longer baselines; SKA-Low has no operating peer at scale (LOFAR is the closest, but smaller). The combination of frequency coverage, survey speed, and sensitivity makes SKA qualitatively rather than incrementally different from existing facilities.

Why now

The radio-astronomy survey case for SKA was made in the 1990s; the technological case became affordable only in the 2010s. Three threads had to converge:

• Affordable fibre and silicon. Aperture arrays at SKA-Low's scale are computationally bottlenecked — each station beam-forms 256 dual-polarisation streams in real time, every station, every observation. The arithmetic became affordable with cheap GPUs and ASICs in the 2010s.
• Calibration breakthroughs. Removing direction-dependent ionospheric effects from 100 MHz observations is genuinely hard. Pipelines developed at LOFAR and MWA in the 2010s — direction-dependent self-calibration, peeling, demixing — are the reason SKA-Low has a realistic shot at the 21 cm signal.
• Site stewardship. Both host nations passed radio-quiet-zone legislation early enough that the sites are still protected as terrestrial transmitter density continues to rise. Without those reserves, the project would be foreclosed by 4G, 5G, low-Earth-orbit satellite constellations, and the general rise of the electromagnetic background.

Open risks

• Satellite constellations. Starlink and OneWeb constellations transmit downlink in bands adjacent to SKA-Mid. Mitigation is by data flagging and by negotiation with operators on transmission patterns; whether this is sufficient at scale is an active problem.
• 21 cm foreground subtraction. Foregrounds (galactic synchrotron, extragalactic point sources) are 10⁴ to 10⁵ times brighter than the reionization signal. The technique relies on the spectral smoothness of foregrounds vs. the spatial-spectral structure of the signal. This works in simulation; the real instrument has to deliver it.
• Compute scaling. The Science Data Processor's compute budget is fixed; if calibration turns out harder than current estimates (e.g. because the ionosphere is worse than modelled or radio-frequency interference is more pervasive), the SDP becomes the bottleneck.
• Long-baseline phase II. Phase 2 — extending baselines to intercontinental scales by linking SKA dishes to far-flung outstations — is currently aspirational; whether it gets funded is a 2030s question.

Common pitfalls in talking about SKA

• "It's a single telescope." Strictly true at the data-correlation level, but the two halves observe non-overlapping frequency bands. SKA-Low and SKA-Mid don't generally observe the same target at the same time — they cover different frequency windows of different science cases.
• "It will be a square kilometre of dishes." The collecting area target is aggregate aperture, not built dish area. SKA-Low's contribution comes from dipoles, not dishes. The "square kilometre" is a hand-waved aggregate effective area, not a contiguous reflecting surface.
• "It replaces existing telescopes." It augments them. JVLA, ALMA, LOFAR, ngVLA all do things SKA doesn't — and SKA's strength is survey speed and sensitivity in its specific bands, not pixel-scale resolution at every frequency.
• "It's done in 2030." Construction is largely complete by 2030; calibration and commissioning continue well past. The instrument has a 50-year operations horizon — most of its science output is in the 2030s and 2040s.
• "€2 billion is the lifetime cost." No — €2 billion is the construction cost. Operations are roughly €100 million per year times a 50-year lifetime; total lifecycle cost is closer to €7 billion in 2021 currency.