| Abstract : | The Epochs of Reionization and Cosmic Dawn are two crucial epochs in the structure formation of the Universe's first billion years. Detection of redshifted HI 21-cm signal is one of the key scientific goals of the ongoing/upcoming radio telescopes like EDGES, SARAS, MWA, SKA, and a potential probe for exploring these epochs. The HI 21cm signal can be measured using an interferometer or by averaging across the whole sky using a single radio telescope. However, these experiments suffer from systematics and are heavily dependent on the accuracy of foreground removal. Therefore, it is essential to understand the significance of each contaminating component when using non-parametric methods to detect the signal, such as machine learning or Bayesian statistics. Our study uses Artificial Neural Networks (ANNs) to separate the HI power spectrum and associated parameters from the given total observed sky power spectrum, which contains the HI signal, real observed foreground, and systematics. This trained model can assist us in producing better findings from ground-based data for the upcoming radio interferometric experiments like SKA, MWA, and HERA. Our initial findings demonstrate a substantial accuracy in retrieving the signal and their parameters from Cosmic Dawn (CD) and Epoch of Reionization (EoR) utilizing the mock data sets of the signals that include the real observed foreground. |
