Speaker
Description
The Cosmic Dawn (CD) and the Epoch of Reionization (EoR) mark pivotal stages in the early evolution of the Universe, occurring within the first billion years after the Big Bang. Despite their importance, the physical properties of the intergalactic medium (IGM) during these epochs remain poorly constrained by observations. Current and forthcoming low-frequency radio experiments, such as EDGES, SARAS, MWA, and the SKA, aim to detect the redshifted 21-cm signal from neutral hydrogen. However, these efforts are challenged by severe foreground contamination, instrumental systematics, and the complexity of accurate foreground removal. Further complications arise from the Earth’s ionosphere, which introduces frequency-dependent distortions and beam chromatic effects that significantly impede the detection of the global 21-cm signal. Accurate modelling of these ionospheric effects requires accounting for processes such as refraction, absorption, and thermal emission. To overcome these challenges, it is crucial to assess the impact of each source of corruption when applying non-parametric signal-recovery methods. In this work, we are developing a robust machine-learning–based regression framework to recover global 21-cm signal parameters from sky-averaged observations that include contributions from astrophysical foregrounds, ionospheric distortions, and instrumental beam chromaticity. The framework also emphasizes the identification of optimal machine-learning models based on a balance between computational efficiency and predictive performance. Overall, this approach shows strong potential to improve ground-based global 21-cm experiments by enabling reliable signal recovery across both the Cosmic Dawn and the Epoch of Reionization, thereby offering new insights into the early Universe.
