SKA-Low AA*: Early Array for EoR Science

• Initial SKA-Low Configuration (AA*) is a transitional low-frequency aperture array with 307 stations, validated by the AAVS2 prototype to bridge early science and full deployment.
• It features 256 dual-polarized antennas per station, advanced digital beamforming, and robust calibration techniques that support high-quality imaging across 50–350 MHz.
• Performance metrics meet and often exceed sensitivity requirements, while challenges such as spectral chromaticity and radio-frequency interference drive ongoing improvements in element design and calibration.

The Initial SKA-Low Configuration (AA*) is the early operational form of the SKA1-Low low-frequency aperture array. In later forecasting work, AA* is treated as a sub-array of the full SKA1-Low deployment, with 307 stations and a maximum baseline of approximately 74 km, while the full “AA4” configuration has 512 stations. Its empirical foundation is the full-size prototype-station programme, especially AAVS2, which was deployed at the Murchison Radio-astronomy Observatory as “the last full-size engineering prototype station” before construction and used to validate the production station concept in situ (Macario et al., 2021, Pal et al., 11 May 2026, Acharya et al., 26 Jun 2026).

1. Definition, nomenclature, and engineering lineage

Within the SKA-Low development sequence, the fundamental hardware unit is the station: a phased aperture array whose digital beams feed the interferometer. AAVS2 was built to be, for practical RF and system-engineering purposes, a production-like SKA1-Low station, with 256 elements, SKALA4.1 dual-polarisation log-periodic antennas, a station diameter of approximately 40 m, a pseudorandom layout philosophy, and the same signal-chain concept of 16 tiles, fibre transport to the central building, and FPGA-based digital processing. The full SKA1-Low design used in the same study comprises 512 stations, each with 256 antennas, for 131072 antennas in total, with about 50% in a dense core and the remainder on spiral arms out to 65 km (Macario et al., 2021).

The label “AA*” is not used uniformly across the literature. Some later EoR and Cosmic Dawn forecasting papers use AA* explicitly for the initial SKA-Low configuration, while other publications discuss the full Phase-1 baseline or refer instead to AAVS2, AA0.5, or the baseline SKA1-Low design. A plausible implication is that “AA*” functions less as a single frozen hardware name than as a bridge between prototype-station validation and early science operations. In that bridge, AAVS2 leads directly to AA0.5, described as “6 full SKA-Low stations like AAVS2,” and then to the full array (Macario et al., 2021, Pal et al., 11 May 2026).

A later review of high-redshift 21 cm inference makes this distinction explicit by adopting two effective layouts: AA* with 307 stations and AA4 with 512 stations. In that treatment, AA* is the baseline early-science array for inference forecasts, whereas AA4 is the nominal full SKA1-Low deployment (Acharya et al., 26 Jun 2026).

2. Station realization and digital architecture

The station concept underlying AA* is a 256-element, dual-polarised aperture array operating over 50–350 MHz. In AAVS2, the elements are SKALA4.1 aluminium log-periodic antennas with dipoles oriented N–S and E–W, mounted on a wire ground mesh and distributed pseudorandomly within a circular region of approximately 40 m diameter. The 256 antennas are grouped into 16 signal-transport tiles, each tile served by a SMART box for RF-to-optical conversion, aggregated at a Field Node Distribution Hub, transported over approximately 5.5 km of fibre to the control building, and processed by 16 Tile Processing Modules (Macario et al., 2021).

The broader SKA-Low receiver architecture follows the same station logic. The digital chain described for SKA1-Low comprises 512 stations, each with 256 dual-polarisation elements arranged as 16 tiles of 16 antennas. Every element is digitized independently; there is no analog tile beamforming. Each tile feeds a Tile Processing Module, coarse channelization is performed in the frequency domain, per-element 2×2 complex calibration matrices are applied, partial tile beams are formed, and a travelling-sum network produces the station beams forwarded to the central processor. The processed RF band is 50–350 MHz, sampled at 800 MSps, with a 1024-channel oversampled polyphase filterbank yielding 512 coarse complex channels at 0.781 MHz spacing (S. et al., 2023).

The element technology selected for this station concept is SKALA, a dual-polarized active log-periodic dipole array. Its design requirements included a single element covering the full 50–350 MHz band, a field of view of at least $\pm 45^\circ$ from zenith, high sensitivity across the band, a footprint compatible with approximately 1.5 m minimum spacing, and polarization purity characterized with IXR. For the SKALA design over an infinite metallic ground plane, simulations show IXR $\ge 15$ dB across most of the field of view and the 50–450 MHz band, with the ground screen motivated in part by improved low-frequency polarization performance (Acedo et al., 2015).

3. Calibration, beam formation, and imaging behaviour

AAVS2 provides the clearest empirical account of how the initial station concept is calibrated. For all commissioning frequencies except 55 MHz, calibration is Sun-based: observations of 22–24 h are acquired with 0.28 s snapshots, snapshots near solar transit are selected, visibilities are rotated to the Solar position, and complex bandpass gains are derived with MIRIAD’s MFCAL using a 0.14 s solution interval. Short baselines are excluded from the calibration model to reduce diffuse Galactic emission, and the resulting gain solutions are applied to all snapshots. The Sun is treated as a point source with a known spectrum, and the absolute flux density scale is set from quiet-Sun measurements tabulated from Benz (2009). The station primary beam is not included in the gain solve; instead, a post-hoc primary-beam correction is applied using the average embedded element pattern (Macario et al., 2021).

At 55 MHz, the diffuse Galactic background makes the Sun too faint for standard complex-gain calibration. In that case only delay corrections are applied, all-sky images are formed, the Solar peak brightness is measured over about an hour around transit, and a single scaling factor—derived from the expected quiet-Sun flux divided by the measured flux—is used to place all 55 MHz quantities on the absolute scale. This demonstrates that first-order equalisation and delay correction are sufficient for coherent imaging even at the bottom of the band (Macario et al., 2021).

The resulting station imaging behaviour is well characterized. Using AAVS2 as a standalone interferometer with all 256 antennas correlated, all-sky images were formed with natural weighting and 21 arcmin pixels. Solar-transit images show the Sun as the dominant point source at 110 and 320 MHz, while all-sky snapshots near Centaurus A transit at 55, 70, and 110 MHz show the Galactic plane and bright A-team sources at the expected positions. The synthesized beam widths listed for the six commissioning frequencies are $8.0^\circ \times 7.9^\circ$ at 55 MHz, $6.3^\circ \times 6.1^\circ$ at 70 MHz, $4.0^\circ \times 3.9^\circ$ at 110 MHz, $2.8^\circ \times 2.7^\circ$ at 160 MHz, $1.9^\circ \times 1.9^\circ$ at 230 MHz, and $1.4^\circ \times 1.3^\circ$ at 320 MHz. The paper states that “the quality of the obtained images remains typically good at all the analyzed frequencies across 24 hours,” indicating relative stability over long timescales (Macario et al., 2021).

4. Sensitivity, direction dependence, and stability

For AA* and its station prototypes, sensitivity is inherently direction- and time-dependent because the system temperature is set by the station beam convolved with the rotating low-frequency sky. In the AAVS2 analysis this is expressed through the antenna temperature

$$T_{\rm ant}(\nu) = \frac{\int_{4\pi} P_p(\nu,\theta,\phi)\, T(\nu,\theta,\phi)\, d\Omega}{\int_{4\pi} P_p(\nu,\theta,\phi)\, d\Omega},$$

where $P_p$ is the embedded element pattern and $\ge 15$0 is the sky brightness temperature. Station sensitivity is estimated with difference imaging, converted to station SEFD, and then scaled to the full SKA-Low array under the assumption of 512 identical stations through

$\ge 15$1

Measured AAVS2 sensitivities agree with embedded-element-pattern electromagnetic simulations within $\ge 15$2 at all six commissioning frequencies, and the derived SKA-Low sensitivities exceed the requirements at all of them, by up to a factor of $\ge 15$3 (Macario et al., 2021).

The numerical comparison reported for LST 0–8 h is:

Frequency (MHz)	$\ge 15$4 (m²/K)	Requirement (m²/K)
55	150	64
70	210	141
110	630	531
160	810	610
230	700	608
320	810	549

These values are explicitly described as averages over a relatively cold-sky LST interval rather than peak sensitivities, and the paper treats them as conservative because they come from real images that may include calibration imperfections or residual RFI (Macario et al., 2021).

The broader SKA-Low sensitivity literature reinforces that no single scalar sensitivity characterizes the array. Station sensitivity depends strongly on pointing direction, local sidereal time, and polarization. A sensitivity calculator based on precomputed values for AAVS2-like and EDA2-like stations stores X, Y, and Stokes I sensitivities in 10 MHz frequency steps, 0.5 h LST intervals, and 5 degree pointing increments; measured data from both stations were used to validate these predictions, and the same framework was verified observationally with pulsar measurements between 70 and 350 MHz (Sokolowski et al., 2022, Lee et al., 2022).

Operational stability is similarly favorable. In AAVS2, a single-snapshot Solar calibration at transit can be transferred across 22–24 h with good imaging quality; self-calibration agrees with single-snapshot calibration up to 160 MHz, showing that gains are stable over several hours; and flux-scale checks on Centaurus A, Fornax A, and Pictor A are consistent with literature spectra, with variations typically within 20% over about an hour around transit (Macario et al., 2021).

5. AA* as an early-science EoR and Cosmic Dawn array

Later end-to-end simulation work treats AA* as a concrete early-science SKA1-Low array with 307 stations and a maximum baseline of approximately 74 km. In a realistic 4-hour tracking simulation at 142 MHz, this configuration is combined with extragalactic point sources, the primary beam response, residual antenna-based gain calibration errors, residual ionospheric phase errors, partial de-mixing of out-of-field sources, and instrumental noise corresponding to 1000 hours of deep integration. Within that setup, Bayesian Gaussian Process Regression recovers the 21 cm signal within the $\ge 15$5 credible interval for almost all k-modes over $\ge 15$6, provided that the covariance model includes baseline-dependent mode mixing and a realistic signal kernel (Pal et al., 11 May 2026).

A complementary review of high-redshift 21 cm inference also uses AA* as the initial SKA-Low configuration and compares it with the full AA4 layout. For power-spectrum forecasts at $\ge 15$7, AA* is evaluated for 100 h and 1000 h integrations; for Bayesian parameter inference, AA* mock observations of 108 h and 1080 h are used. The chapter concludes that AA* can already deliver early science: the midpoint of reionization can be recovered with $\ge 15$8 at 108 h and approximately 0.02 at 1080 h, and the optical depth $\ge 15$9 improves substantially over current constraints. At the same time, the chapter emphasizes that the power spectrum alone does not fully determine all astrophysical parameters, and that combining the power spectrum with higher-order and morphological statistics is important for breaking degeneracies (Acharya et al., 26 Jun 2026).

A plausible implication is that AA* is scientifically defined not only by partial collecting area, but by the fact that it already operates in the regime where calibration residuals, beam chromaticity, and statistical compression become the limiting factors rather than raw detectability alone. That interpretation is consistent with the explicit finding that baseline-dependent mode mixing must be modeled for AA* if wedge leakage is to be controlled in EoR analyses (Pal et al., 11 May 2026).

6. Limitations, extensions, and open issues

Several limitations of the initial configuration are explicit in the literature. First, terminology is not fully standardized: some publications use “AA*” explicitly for the initial 307-station configuration, whereas others discuss the Phase-1 baseline without the AA* label or refer instead to AAVS2 and AA0.5. Second, the initial SKA-Low baseline, even in full Phase 1, is not designed for sub-arcsecond low-frequency imaging. A recent VLBI chapter states that SKA-Low alone has a maximum baseline of 73.4 km and an angular resolution of around 6 arcseconds at 150 MHz; to reach better than 0.1 arcseconds at 150 MHz, it proposes LAMBDA, a network of 4–6 SKA-Low-like stations distributed throughout Australia to provide baselines of up to approximately 4000 km (Timmerman et al., 23 Jun 2026).

Third, spectral smoothness remains a design-sensitive issue. A 2025 study of SKA-Low station chromaticity finds that pseudo-random layouts such as AAVS2 and the first deployed S8 stations are much better than regular layouts such as AAVS3 for suppressing extra spectral structure from mutual coupling, but it also concludes that the dominant chromatic effects come from the SKALA4.1 reflection coefficient and the finite ground screen rather than from station geometry alone. In that analysis, realistic pseudo-random stations still show 20–40% amplitude and 10–20° phase structure across the band once element and ground effects are included, and the paper argues that improving the initial SKA-Low configuration for Cosmic Dawn and EoR is therefore more about element design, ground-screen behaviour, and beamforming or apodisation than about minor layout changes (Staveley-Smith, 24 Mar 2025).

Fourth, the low-frequency satellite-RFI environment is already visible with station analogs. Using EDA2, a 256-element, 35 m SKA-Low station analog at the SKA-Low site, strong intended and unintended emissions from Starlink satellites were detected at 137.5 and 159.4 MHz with 0.926 MHz bandwidth and 2 s imaging. The brightest detected transmissions reach $8.0^\circ \times 7.9^\circ$0 Jy/beam, and the paper concludes that both intended and unintended radiation from Starlink satellites will be detrimental to key SKA science goals without mitigation (Grigg et al., 2023).

Taken together, these results define the initial SKA-Low configuration as an already capable but still transitional instrument: empirically validated at the station level, analytically tractable enough for early EoR and Cosmic Dawn inference, yet still limited by chromaticity, calibration complexity, nomenclature ambiguity, and the practical realities of the low-frequency interference environment.