GLOBALEMU: A novel and robust approach for emulating the sky-averaged 21-cm signal from the cosmic dawn and epoch of reionisation

Abstract

Emulation of the Global (sky-averaged) 21-cm signal with neural networks has been shown to be an essential tool for physical signal modelling. In this paper we present globalemu, a Global 21-cm signal emulator that uses redshift as a character defining variable alongside a set of astrophysical parameters to estimate the signal brightness temperature. Combined with physically-motivated data pre-processing this makes for a reliable and fast emulator that is relatively insensitive to the network design. globalemu can emulate a high resolution signal in 1.3 ms in comparison to 133 ms, a factor of 102 improvement, when using the existing public state of the art 21cmGEM. We illustrate, with the standard astrophysical models used to train 21cmGEM, that globalemu is almost twice as accurate and for a test set of $\approx1,700$ signals we achieve a mean RMSE of 2.52 mK across the band z=7-28 ($\approx$10 per cent the expected noise of the Radio Experiment for the Analysis of Cosmic Hydrogen (REACH)). The models are parameterised by the star formation efficiency, $f_*$, minimum virial circular velocity, $V_c$, X-ray efficiency, $f_X$, CMB optical depth, $\tau$, the slope and low energy cut off of the X-ray spectral energy density, $\alpha$ and $\nu_\mathrm{min}$, and the mean free path of ionizing photons, $R_\mathrm{mfp}$. globalemu provides a flexible framework for easily emulating updated simulations of the Global signal and in addition the neutral fraction history. The emulator is pip installable and available at: \url{https://github.com/htjb/globalemu}. globalemu will be used extensively by the REACH collaboration.