🔭 astrophysics

SKA-Low simulations for a cosmic dawn/epoch of reionisation deep field

This paper presents a realistic 1000-hour SKA-Low simulation of the cosmic dawn and epoch of reionisation, complete with complex foregrounds, instrumental errors, and the true underlying signal, to serve as a benchmark for developing and testing foreground-mitigation techniques, notably for the SKA Science Data Challenge 3a.

Original authors: Anna Bonaldi, Philippa Hartley, Simon Purser, Omkar Bait, Eunseong Lee, Robert Braun, Florent Mertens, Andrea Bracco, Wendy Williams, Cath Trott

Published 2026-02-18

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Anna Bonaldi, Philippa Hartley, Simon Purser, Omkar Bait, Eunseong Lee, Robert Braun, Florent Mertens, Andrea Bracco, Wendy Williams, Cath Trott

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine trying to hear a single, tiny whisper from a newborn baby (the Cosmic Dawn) in the middle of a roaring, chaotic stadium filled with thousands of screaming fans, a marching band, and a giant PA system blasting static (the Foregrounds).

That is essentially what astronomers are trying to do with the Square Kilometre Array (SKA), a massive radio telescope project. They want to listen to the "baby" of the universe—the moment when the first stars turned on and began to ionize the gas around them. But the "crowd" of noise is billions of times louder than the signal they are looking for.

This paper is about building a hyper-realistic "flight simulator" for this specific challenge. Instead of just guessing how to filter out the noise, the authors created a perfect digital copy of what the telescope should see, including all the messy errors and interference, so they can practice cleaning up the data before the real telescope even starts listening.

Here is a breakdown of the paper using everyday analogies:

1. The Goal: Finding the "Ghost" Signal

The universe started as a dark, neutral fog. Then, the first stars ignited, creating bubbles of ionized gas. This transition is called the Epoch of Reionisation (EoR). Astronomers want to map these bubbles using a specific radio frequency (the 21cm line from hydrogen).

The Problem: The signal from these ancient bubbles is incredibly faint. Meanwhile, our own galaxy (the Milky Way) and distant radio galaxies are screaming at the telescope with radio waves that are 1,000 to 10,000 times brighter than the signal they want.
The Analogy: It's like trying to hear a mosquito buzzing in a hurricane. The hurricane (foregrounds) isn't just loud; it's also changing shape and direction, making it hard to tell where the mosquito is.

2. The Simulator: Building a "Digital Twin"

The authors didn't just make a simple model; they built a digital twin of a 1,000-hour observation session.

The Telescope: They simulated the SKA-Low array, which consists of 512 "stations" (groups of antennas) spread out over a huge area.
The Sky: They filled their digital sky with:
- The "Whisper": A realistic simulation of the Cosmic Dawn signal.
- The "Crowd": Millions of real and simulated radio galaxies, some so bright they are visible even when they are far outside the telescope's direct view (like hearing a siren from a street away).
- The "Static": The Milky Way's own radio glow, but with added fine details (like the texture of a fabric) that standard maps usually miss.

3. The "Glitches": Why Real Life is Hard

The most important part of this paper is that they didn't just simulate a perfect world. They intentionally broke things to mimic reality.

The Ionosphere (The Shimmering Heat Haze): Just as heat waves from a road make a car look like it's wobbling, the Earth's ionosphere distorts radio waves. The authors simulated this "shimmering" effect, which scrambles the data.
Calibration Errors (The Tuning Knob): Real telescopes aren't perfect. The sensitivity of the antennas changes slightly over time and frequency. The authors added "jitter" to the volume and timing of the signal to mimic these imperfections.
The "Leakage" (The Bad Neighbors): Because the telescope has "side lobes" (like a flashlight that isn't perfectly focused), bright sources outside the main view can leak into the image. The authors simulated a process where they try to subtract these leaks, but admit they can't get rid of them perfectly.

4. The Challenge: The "Clean Up" Game

The authors released this simulation (along with the code used to make it) as a benchmark challenge (called SDC3a).

The Game: They gave the data to other scientists and said, "Here is the 'dirty' picture with all the noise and glitches. Can you use your algorithms to clean it up and find the 'whisper' (the EoR signal) hidden inside?"
The Truth: Crucially, they also provided the "True Image" (the clean signal without any noise). This allows them to grade the scientists' work: "You got 80% of the signal back, but you accidentally erased some of the baby's face."

5. The Results: What Did They Learn?

When scientists tried to clean this simulated data, they found that:

It's a multi-step dance: You can't just use one filter. You have to remove the loudest "screaming fans" (bright galaxies) first, then smooth out the "marching band" (diffuse galaxy glow), and finally deal with the "static" (instrumental errors).
Errors are tricky: If you don't account for the ionosphere and calibration errors perfectly, your cleaning tools might accidentally think the "whisper" is just more noise and delete it.
The "Wedge": The simulation showed that instrumental errors cause the bright foreground noise to "leak" into the quiet zones where the EoR signal lives, making it much harder to detect.

Summary

This paper is a training manual for the future. By creating a perfect, messy, realistic simulation of what the SKA telescope will see, the authors are giving the scientific community a safe place to practice their "noise-canceling" techniques.

Think of it as a flight simulator for radio astronomers. Before they fly the real plane (the actual SKA telescope) into the stormy skies of the early universe, they need to crash a few times in the simulator to learn how to steer through the turbulence and finally hear the first whispers of the universe's dawn.

1. Problem Statement

The detection of the 21-cm signal from neutral hydrogen (HI) during the Cosmic Dawn (CD) and Epoch of Reionisation (EoR) is a primary science goal for the Square Kilometre Array Observatory (SKAO). However, this signal is extremely faint (micro-Kelvin level) and is obscured by foreground emissions that are 3–4 orders of magnitude brighter. These foregrounds include:

Galactic emission: Synchrotron and free-free radiation from the Milky Way.
Extragalactic emission: Point sources and diffuse emission from other galaxies.
Instrumental and Environmental Effects: The chromatic nature of telescope beams, ionospheric fluctuations, Radio Frequency Interference (RFI), and calibration errors (both Direction-Dependent and Direction-Independent).

Existing simulations often lack the realism required to test foreground mitigation techniques against the complex interplay of these factors. There is a critical need for a high-fidelity simulation that incorporates realistic sky models, instrumental responses, and error sources to benchmark data analysis pipelines before the SKA begins operations.

2. Methodology

The authors generated a comprehensive simulation of a deep-field observation using the OSKAR (Oxford SKA Array) package. The simulation mimics a 1000-hour integration of a single pointing over the frequency range 106–196 MHz (corresponding to redshifts $z \approx 6.2–12.4$ ).

A. Telescope Model

Configuration: Modeled after the full SKA-Low array with 512 stations using the "Vogel" spiral layout.
Observing Strategy: A 4-hour track was simulated and scaled to represent a 1000-hour deep integration.
Simplifications: To manage computational cost, a single station layout was used for all 512 stations (rather than 512 unique layouts), and a single polarization state was assumed. This results in elevated far-side-lobe responses compared to the final deployed array.

B. Sky Model Components

The sky model was constructed to be a plausible realization of the true sky, not a prediction of the specific field, to ensure a "blind" test for mitigation algorithms.

Cosmic Dawn/EoR Signal: Generated using 21cmFAST (v3.2.0) with Planck18 cosmology. The signal was converted to observational units and added to a coarse grid.
Extragalactic Sources:
- Out-of-field: Sources brighter than 5 Jy at 150 MHz (approx. 1200 sources) and "A-Team" sources were included over the full $2\pi$ steradians.
- In-field: Sources down to 1 $\mu$ Jy at 150 MHz were included over an $8 \times 8$ degree field. Strong sources (>100 mJy) were individual components; fainter sources were modeled using the T-RECS simulation, represented as a mix of resolved galaxies, AGN, and unresolved sources.
- Total Count: Over 15 million sources were modeled.
Galactic Diffuse Emission: Based on the GSM2016 model, supplemented with small-scale structure (<1 degree) derived from a Magneto-Hydrodynamic (MHD) simulation of the interstellar medium to capture filamentary structures.
Grid Implementation: To balance accuracy and computational load, sources were mapped onto two grids: a "Fine grid" (5760 $\times$ 5760 pixels, 5 $\times$ 5 arcsec) and a "Coarse grid" (512 $\times$ 512 pixels, 56.25 $\times$ 56.25 arcsec).

C. Error Model

The simulation introduces realistic corrupting effects to test mitigation robustness:

Partial De-mixing of Out-of-field Sources: Instead of a full de-mixing pipeline, the authors applied an "effective" primary beam attenuation ( $10^{-3}$ factor) to simulate the residual impact of strong sources entering via side-lobes after partial subtraction.
Ionospheric Effects: Modeled using ARatmospy to create a time-evolving phase screen. The simulation assumes a correlation scale of $r_0 = 7$ km and applies a $10^{-2}$ attenuation factor to simulate the improvement expected from Direction-Dependent (DD) calibration over 250 tracks.
Calibration Errors:
- Direction-Independent (DI): Gaussian noise colored to have a $-2$ power-law index (red noise) in both time and frequency domains. Residual errors were set to 0.02 degrees (phase) and 0.02% (amplitude) after averaging.
- Direction-Dependent (DD): Included via the ionospheric phase screen.
Thermal Noise: Scaled to represent the sensitivity of a 1000-hour integration.

3. Key Contributions

Benchmark Dataset: The creation of the most sophisticated SKA-Low EoR simulation to date, serving as the basis for the SKA Science Data Challenge 3a (SDC3a).
Open Source Framework: The release of the code repositories used to generate the simulation, allowing the community to produce new datasets with varying parameters.
Multi-Component Realism: Unlike previous simulations, this work integrates:
- Diffuse and point-source foregrounds simultaneously.
- Small-scale Galactic structure via MHD simulations.
- A realistic mix of instrumental errors (DD/DI calibration, ionosphere, side-lobe leakage).
Data Products: Delivery of both Visibilities (7.5 TB in Measurement Set format) and Imaging products (natural weighting, no deconvolution), alongside the "truth" noiseless EoR signal for validation.

4. Results and Analysis

Power Spectrum Analysis: The authors analyzed the 2D power spectra ( $k_\parallel$ $k_{∥}$ vs. $k_\perp$ $k_{⊥}$ ) of the components.
- Galactic Emission: Dominates at low $k_\parallel$ but leaves a relatively clean "window" for EoR detection if instrumental effects are ignored.
- Instrumental Effects: The addition of ionospheric distortions and calibration errors causes "mode leakage," contaminating the clean window and pushing foreground power into the EoR signal region.
- Extragalactic Sources: Introduce contamination at larger $k_\parallel$ .
SDC3a Insights: The simulation was used in the SDC3a challenge, which revealed that:
- Effective decoupling of complexity requires multiple analysis stages (e.g., iterative subtraction of strong point sources before diffuse foreground removal).
- Explicit deconvolution/convolution steps to restore spectral smoothness yield the best results.
- Current error estimation methods are not robust enough against residual systematics after foreground mitigation.

5. Significance

This paper represents a significant step forward in preparing for the SKA era. By providing a "truth" dataset that includes the complex, non-linear interactions between sky signals and instrumental errors, it allows researchers to:

Validate Algorithms: Rigorously test foreground mitigation techniques (subtraction vs. avoidance) under realistic conditions.
Quantify Systematics: Understand how calibration errors and ionospheric effects degrade the EoR signal recovery.
Guide Future Observations: Inform the design of observation strategies and calibration pipelines for the upcoming SKA-Low telescope.

The simulation framework and data products are publicly available, establishing a standard benchmark for the global EoR community to refine their analysis pipelines before the arrival of real SKA data.