Extracting the 21-cm Power Spectrum and the reionization parameters from mock datasets using Artificial Neural Networks
Detection of the \hi~ 21-cm power spectrum is one of the key science drivers of several ongoing and upcoming low-frequency radio interferometers. However, the major challenge in such observations come from bright foregrounds, whose accurate removal or avoidance is key to the success of these experiments. In this work, we demonstrate the use of artificial neural networks (ANNs) to extract the \hi~ 21-cm power spectrum from synthetic datasets and extract the reionization parameters from the \hi~ 21-cm power spectrum. For the first time, using a suite of simulations, we present an ANN based framework capable of extracting the \hi~ signal power spectrum directly from the total observed sky power spectrum (which contains the 21-cm signal, along with the foregrounds and effects of the instrument). To achieve this, we have used a combination of two separate neural networks sequentially. As the first step, \texttt{ANN1} predicts the 21-cm power spectrum directly from foreground corrupted synthetic datasets. In the second step, \texttt{ANN2} predicts the reionization parameters from the predicted \hi~ power spectra from \texttt{ANN1}. Our ANN-based framework is trained at a redshift of $9.01$, and for \kk-modes in the range, $\rm{0.17<\kk<0.37~Mpc^{-1}}$. We have tested the network's performance with mock datasets that include foregrounds and are corrupted with thermal noise, corresponding to $1080$ hrs of observations of the \textsc{ska-1 low} and \textsc{hera}. Using our ANN framework, we are able to recover the \hi~ power spectra with an accuracy of $\approx95-99\%$ for the different test sets. For the predicted astrophysical parameters, we have achieved an accuracy of $\approx~81-90\%$ and $\approx~50-60\%$ for the test sets corrupted with thermal noise corresponding to the \textsc{ska-1 low} and \textsc{hera}, respectively.