Needlets and foreground removal for SKAO hydrogen intensity maps

TL;DR

This work addresses the challenge of recovering the faint 21-cm HI signal in SKA-like single-dish intensity mapping by applying a needlet-based PCA (Need-PCA) and benchmarking it against GMCA and GNILC. By leveraging spherical needlets, the authors achieve double localization in real and harmonic space and test the pipeline on realistic simulations that include HI, astrophysical foregrounds, polarization leakage, and beam effects. Across 0.41 ≤ z ≤ 0.58 and angular scales 30 ≤ ℓ ≤ 136, all methods recover the HI power spectrum within about 10% accuracy, with Need-PCA/Need-GMCA showing robustness to beam sidelobes and masking; GNILC tends to over-clean at large scales. The results support the viability of needlet-based foreground separation for SKAO-MID-like HI intensity mapping and highlight its resilience to systematics, guiding future data-analysis pipelines.

Abstract

Intensity Mapping (IM) of the 21-cm line of the neutral hydrogen (\textsc{Hi}) has become a compelling new technique to map the large-scale structure of the Universe. One of the main challenges is the presence of strong foreground emissions of several orders of magnitude larger than the \textsc{Hi}~signal. Here, we implement a version of the Principal Component Analysis, a blind component-separation technique, based on a kind of spherical wavelets called needlets. These functions exploit double localization both in real and in harmonic space. We test Need-PCA performances on a set of maps that simulates the SKA MID radio telescope in the AA4 configuration. We compare our results with other component separation methods such as Generalised Morphological Component Analysis (GMCA) and Generalized Needlet Internal Linear Combination (GNILC). All the methods have comparable results, recovering the \textsc{Hi}~signal within 10\% accuracy across the frequency channels, in the multipole range 30 $\lesssim \ell \lesssim$ 136. We also test our pipeline in the presence of systematics such as polarization leakage. We find that the cleaning methods are insensitive to the presence of such systematic, yielding the same results as in the leakage-free case. Finally, under the assumption of a realistic telescope beam with sidelobes, we find that standard PCA and GMCA fails to recover the \textsc{Hi}~signal at larger scales, while the Need-PCA and Need-GMCA are less affected. GNILC tends to over-clean, yielding to a loss of the signal.

Figures (20)

• Figure 1: Mollweide projection of the intensity of the foreground and systematic contaminants, in the frequency channel 952.5 MHz. From top left and clockwise, the maps show: galactic synchrotron, galactic free-free, polarization leakage and extragalactic point sources. The units are in mK, the scale is logarithmic for the synchrotron and free-free maps and linear for the point sources and the polarization leakage.
• Figure 2: Brightness temperature as a function of the frequency along a line of sight, in the direction of lon=0 deg, lat=-5 deg.
• Figure 3: Total temperature map of the observed dataset at channel $\nu=952.5$ MHz. The applied mask covers the 50% of the pixels in the Galactic plane.
• Figure 4: Power spectrum of the simulated components as a function of the angular scales. Left-Top panel: power spectra at the frequency $\nu=$952 MHz. Right-Top panel: power spectra in the frequency range $\nu \in \left[ 900.5 - 1004.5 \right]$ MHz after the convolution with the telescope beam. Bottom panel: angular power spectra of the contaminants when a mask is applied, covering the 50% of the sky in the Galactic plane.
• Figure 5: Weight functions for needlets with $j_{\rm max}=4$, $\ell_{\rm max}=383$. The gradient of colors changes from green to blue as the band index j increases from $0$ to $4$. The solid line corresponds to the weight function of the standard needlets, the dashed line of the cosine ones.