Title: Separating the EoR Signal with a Convolutional Denoising Autoencoder: A Deep-learning-based Method
Authors: Weitian Li, Haiguang Xu, Zhixian Ma, Ruimin Zhu, Dan Hu, Zhenghao Zhu, Junhua Gu, Chenxi Shan, Jie Zhu, Xiang-Ping Wu
Journal: MNRAS, 2019, 485, 2628
arXiv: 1902.09278
ADS: 2019MNRAS.485.2628L
Code: https://github.com/liweitianux/cdae-eor
Abstract: When applying the foreground removal methods to uncover the faint cosmological signal from the epoch of reionization (EoR), the foreground spectra are assumed to be smooth. However, this assumption can be seriously violated in practice since the unresolved or mis-subtracted foreground sources, which are further complicated by the frequency-dependent beam effects of interferometers, will generate significant fluctuations along the frequency dimension. To address this issue, we propose a novel deep-learning-based method that uses a 9-layer convolutional denoising autoencoder (CDAE) to separate the EoR signal. After being trained on the SKA images simulated with realistic beam effects, the CDAE achieves excellent performance as the mean correlation coefficient (ρ̄) between the reconstructed and input EoR signals reaches 0.929 ± 0.045. In comparison, the two representative traditional methods, namely the polynomial fitting method and the continuous wavelet transform method, both have outstanding difficulties in uncovering the EoR signal, yielding only ρ̄[poly] = 0.296 ± 0.121 and ρ̄[cwt] = 0.198 ± 0.160, respectively. We conclude that, by hierarchically learning sophisticated features through multiple convolutional layers, the CDAE is a powerful tool that can be used to overcome the complicated frequency-dependent beam effects and accurately separate the EoR signal, which exhibits the great potential of deep-learning-based methods in future EoR experiments.