Seeing Wiggles without Seeing Wiggles: BAO Recovery in 21 cm Intensity Mapping with Deep Learning

TL;DR

This work addresses the challenge of losing large-scale information in 21 cm intensity mapping due to the foreground wedge. It introduces a 3D U-Net that reconstructs missing large-scale modes from short-wavelength data, trained exclusively on de-wiggled simulations to ensure recovery arises from non-linear mode coupling rather than memorization of BAO features. The results show high-fidelity amplitude and phase recovery in the noise-free case and robust phase information under realistic noise and cosmological variations, with BAO features recoverable via both direct and template-fitting approaches. This mode-restoration approach offers a promising, complementary method for extracting BAO information in future 21 cm IM analyses, while highlighting areas for improvement such as higher-resolution training data and inclusion of additional systematics.

Abstract

The 21 cm intensity mapping provides a promising probe of the large-scale structure. Astrophysical foregrounds, as the main source of contamination to the cosmological 21 cm signal, persist in a wedge-like region of Fourier space due to the inherent chromaticity in radio interferometric observations. The foreground avoidance strategy focuses on utilizing data from relatively clean regions with minimal foreground leakage, at the cost of losing large-scale information. Non-linear structure formation, however, couples Fourier modes across scales, leaving imprints of the missing large-scale modes in the remaining data. In this work, we employ a deep learning approach to test whether large-scale features of the 21 cm brightness temperature fields, particularly the baryon acoustic oscillations (BAO), can be recovered at the field level using only short-wavelength modes that are beyond the linear scales. To explicitly assess the dependence on the training cosmology, we train the network exclusively on de-wiggled simulations, providing a controlled test of whether the reconstruction arises from physical non-linear mode coupling rather than implicit encoding of BAO features. In the ideal noise-free case, the amplitude and phase of the lost modes can be restored with high fidelity. With instrumental noise included, the reconstructed amplitude becomes biased, while the phase information remains robust. The trained network also exhibits reasonable robustness to variations in the underlying cosmological model. Together, these results suggest that mode restoration offers a complementary approach for extracting cosmological information from future 21 cm intensity mapping analyses.

Figures (10)

• Figure 1: A slice of the dark matter distribution (left) and the corresponding fluctuation of $21\,\mathrm{cm}$ brightness temperature (right), taken in a plane transverse to the line-of-sight direction. The red dots in the left panel indicate the position of halos.
• Figure 2: Left: The configuration of 164 SKA-Mid AA4 antennas located within $10\,$km of the center of array. Right: A portion of the baseline distribution in the uv plane for the SKA-Mid AA4 configuration, assuming an $8\,$h tracking observation of the COSMOS field at $z=1.0$. Owing to the limited spatial resolution of our simulations, only a truncated region of the uv plane, shown as the zoomed-in part area, is retained for the generation of thermal noise.
• Figure 3: Diagram of the neural network architecture. The network follows an encoder-decoder structure in which the fields after removing specific modes are processed through successive 3D convolution layers (black arrows) and down-sampling layers (orange arrows). The decoder employs 3D transposed convolutions layers (red arrows) to perform up-sampling operations, and incorporates skip connections (blue arrows) through concatenation. A global residual connection (green arrow) performs element-wise addition between the input and network output. Before being passed to the network, each datacube is padded from a size of $64^3$ to $104^3$. The network output has a dimension of $64^3$ and therefore requires no additional cropping.
• Figure 4: Slices of the $21\,\mathrm{cm}$ brightness temperature fields with volume of $(1024\, h^{-1} {\rm Mpc})^3$. The top row shows a slice along the direction of line-of-sight, the bottom row shows a slice along a spatial direction. The first column shows the $21\,\mathrm{cm}$ field without removing modes in the foreground wedge and linear scales. The second and third columns show the field after removing the specific modes and the corresponding neural network reconstruction for the noise-free case. The last two columns display the modes removed field and reconstructed field for the case including noise. In each panel, a selected region is zoomed in to more clearly illustrate the reconstruction performance.
• Figure 5: The spherically averaged power spectra (top), transfer function (middle), and cross-correlation coefficient (bottom) for the noise-free (left), and the noisy (right) $21\,\mathrm{cm}$ brightness temperature fields. The vertical dashed lines therein indicate the cutoff scale below which modes with $k < 0.3 \,h{\rm Mpc}^{-1}$ are removed from the fields when performing reconstruction using the network. Specifically, in the top right panel, we additionally show the power spectra of the noise-free input field (orange line) and noise data (gray line).