deep21: a Deep Learning Method for 21cm Foreground Removal

Abstract

We seek to remove foreground contaminants from 21cm intensity mapping observations. We demonstrate that a deep convolutional neural network (CNN) with a UNet architecture and three-dimensional convolutions, trained on simulated observations, can effectively separate frequency and spatial patterns of the cosmic neutral hydrogen (HI) signal from foregrounds in the presence of noise. Cleaned maps recover cosmological clustering statistics within 10% at all relevant angular scales and frequencies. This amounts to a reduction in prediction variance of over an order of magnitude on small angular scales ($\ell >300$), and improved accuracy for small radial scales ($k_{\parallel }>0.17\mathrm{h}\mathrm{Mp}{\mathrm{c}}^{-1})$ compared to standard Principal Component Analysis (PCA) methods. We estimate posterior confidence intervals for the network's prediction by training an ensemble of UNets. Our approach demonstrates the feasibility of analyzing 21cm intensity maps, as opposed to derived summary statistics, for upcoming radio experiments, as long as the simulated foreground model is sufficiently realistic. We provide the code used for this analysis on Github https://github.com/tlmakinen/deep21 as well as a browser-based tutorial for the experiment and UNet model via the accompanying http://bit.ly/deep21-colab Colab notebook.

Published in JCAP 30 April 2021. 30 pages, 11 figures

References in corpus (18)

• Cosmology at Low Frequencies: The 21 cm Transition and the High-Redshift Universe
• Measurement of 21 cm brightness fluctuations at z ~ 0.8 in cross-correlation
• Late-time cosmology with 21cm intensity mapping experiments
• Foreground simulations for the LOFAR - Epoch of Reionization Experiment
• Fast likelihood-free cosmology with neural density estimators and active learning
• A Method for 21cm Power Spectrum Estimation in the Presence of Foregrounds
• Testing General Relativity with 21 cm intensity mapping
• Blind foreground subtraction for intensity mapping experiments
• An Improved Method for 21cm Foreground Removal
• Simulating polarized Galactic synchrotron emission at all frequencies, the Hammurabi code
• Results from EDGES High-Band: I. Constraints on Phenomenological Models for the Global $21$ cm Signal
• Fast simulations for intensity mapping experiments
• Baryonic acoustic oscillations from 21cm intensity mapping: the Square Kilometre Array case
• Impact of Foregrounds on HI Intensity Mapping Cross-Correlations with Optical Surveys
• Separating the EoR Signal with a Convolutional Denoising Autoencoder: A Deep-learning-based Method
• Constraining the reionization history using deep learning from 21cm tomography with the Square Kilometre Array
• A unified framework for 21cm tomography sample generation and parameter inference with Progressively Growing GANs
• Deep-Learning Study of the 21cm Differential Brightness Temperature During the Epoch of Reionization

Cited by in corpus (42)

• SKAO HI Intensity Mapping: Blind Foreground Subtraction Challenge
• The CAMELS Multifield Dataset: Learning the Universe's Fundamental Parameters with Artificial Intelligence
• From a locality-principle for new physics to image features of regular spinning black holes with disks
• The CAMELS project: public data release
• Recovering the Wedge Modes Lost to 21-cm Foregrounds
• Machine learning astrophysics from 21 cm lightcones: impact of network architectures and signal contamination
• Lossless, Scalable Implicit Likelihood Inference for Cosmological Fields
• Deep learning approach for identification of HII regions during reionization in 21-cm observations
• The foreground transfer function for HI intensity mapping signal reconstruction: MeerKLASS and precision cosmology applications
• Machine Learning for Observational Cosmology
• Eliminating Primary Beam Effect in Foreground Subtraction of Neutral Hydrogen Intensity Mapping Survey with Deep Learning
• Detecting the power spectrum turnover with HI intensity mapping
• The local PNG bias of neutral Hydrogen, ${\mathrm{H}}_{\mathrm{I}}$
• Machine-learning recovery of foreground wedge-removed 21-cm light cones for high-$z$ galaxy mapping
• YOLO-CIANNA: Galaxy detection with deep learning in radio data. I. A new YOLO-inspired source detection method applied to the SKAO SDC1
• Improving constraints on primordial non-Gaussianity using neural network based reconstruction
• Eliminating polarization leakage effect for neutral hydrogen intensity mapping with deep learning
• Machine Learning and Cosmology
• Extracting the Signal of Cosmic String Wakes from 21-cm Observations
• 21-cm foreground removal using AI and frequency-difference technique
• Foreground Removal of CO Intensity Mapping Using Deep Learning
• Probing light relics through cosmic dawn
• Deep learning approach for identification of HII regions during reionization in 21-cm observations -- III. image recovery
• Field-level Reconstruction from Foreground-Contaminated 21-cm Maps
• MAGIC: Microlensing Analysis Guided by Intelligent Computation
• Measuring the global 21-cm signal with the MWA-II: improved characterisation of lunar-reflected radio frequency interference
• Perspectives on fundamental cosmology from Low Earth Orbit and the Moon
• Neural Network Reconstruction of Non-Gaussian Initial Conditions from Dark Matter Halos
• Application of Physics-Informed Neural Networks in Removing Telescope Beam Effects
• Map Reconstruction of radio observations with Conditional Invertible Neural Networks
• An Alcock-Paczynski Test on Reionization Bubbles for Cosmology
• Primary Beam Chromaticity in HIRAX: I. Characterization from Simulations and Power Spectrum Implications
• Extracting the Epoch of Reionization Signal with 3D U-Net Neural Networks Using Data-driven Systematic Effect Model
• Squeezing information from radio surveys to probe the primordial Universe
• CMB delensing with deep learning
• Foreground removal in HI 21 cm intensity mapping under frequency-dependent beam distortions
• Efficient neutral-IGM inference from noisy 21-cm forest spectra with latent-space U-Net encoding and XGBoost
• Deep learning with hybrid frequency differencing and principal component analysis for 21-cm foreground and beam mitigation
• Beyond the Power Spectrum: A New Framework for Non-Stationary Fields with Applications to Light-Cone Effects in Line Intensity Mapping
• Simulating a full-sky high resolution Galactic synchrotron spectral index map using neural networks
• Exploring delaying and heating effects on the 21-cm signature of fuzzy dark matter
• Foreground Subtraction with a Tensor-Based Oriented Singular Value Decomposition Method for HI Experiments