Global 21 cm Signal Recovery Under Changing Environmental Conditions

TL;DR

This paper presents a novel method using Masked Auto-regressive Flows to improve the recovery of the 21 cm cosmic signal under changing environmental conditions, outperforming traditional Bayesian approaches.

Contribution

The authors introduce a new tool that enhances 21 cm signal recovery in variable environments, addressing limitations of previous models.

Findings

Improved signal recovery with up to 45% reduction in RMSE.

Higher Bayesian evidence with the new method under changing conditions.

Effective in scenarios with dynamic soil dielectric properties and temperatures.

Abstract

The redshifted 21 cm line of cosmic atomic hydrogen is one of the most auspicious tools in deciphering the early Universe. Recovering this signal remains an ongoing problem for cosmologists in the field, with the signal being hidden behind foregrounds approximately five orders of magnitude brighter than itself. A traditional forward modelling data analysis pipeline using Bayesian data analysis and a physically motivated foreground model to find this signal shows great promise in the case of unchanging environmental conditions. However we demonstrate in this paper that in the presence of a soil with changing dielectric properties under the antenna over time, or a changing soil temperature in the far field of our observation these traditional methods struggle. In this paper we detail a tool using Masked Auto-regressive Flows that improves upon previous physically motivated foreground models when one is trying to recover this signal in the presence of changing environmental conditions. We demonstrate that with these changing parameters our tool consistently recovers the signal with a much greater Bayesian evidence than the traditional data analysis pipeline, decreasing the root mean square error in the recovery of the injected signal by up to 45 %.