Researchers from several institutions, including the University of Science and Technology of China and the Chinese Academy of Sciences, have developed a novel approach to improve the accuracy of 21-cm intensity mapping (IM), a technique used to study the large-scale distribution of neutral hydrogen (HI) and extract cosmological information. Their work, published in the journal Physical Review D, focuses on mitigating the effects of bright foregrounds and frequency-dependent beam effects that can compromise traditional cleaning methods and obscure the faint HI signal.
The team built upon their previous work on the frequency-differencing (FD) method, which helps to separate the cosmological signal from foreground contamination. In this new study, they systematically compared FD-based and principal component analysis (PCA)-based reconstructions using realistic simulations that included both foregrounds and beam convolution. They found that both methods performed similarly well when there was no beam or when a Gaussian beam was used. However, under a more realistic cosine beam, both methods systematically underestimated the large-scale cross-correlation power spectrum, particularly for scales larger than about 10 megaparsecs (a megaparsec is roughly 3.26 million light-years).
To address this limitation, the researchers explored a hybrid approach. They trained a deep-learning model called UNet with two input channels: one constructed from FD and the other from PCA cleaning. This allowed the network to simultaneously exploit the strengths of both inputs. The two-channel strategy achieved superior performance, maintaining the cross-correlation power spectrum close to unity on large scales under a cosine beam. This represents an improvement of 5-8% relative to either FD-based or PCA-based UNet alone.
The researchers suggest that providing complementary FD and PCA information to a single deep network is an effective route to robust HI reconstruction. This advancement could lay the groundwork for precision baryon acoustic oscillation (BAO) measurements with future low-redshift 21 cm IM surveys. BAO measurements are crucial for understanding the expansion history of the universe and the nature of dark energy.
This research highlights the potential of deep-learning techniques to enhance the analysis of cosmological data, offering more accurate and reliable methods for studying the universe’s large-scale structure. For the energy sector, improved cosmological measurements can indirectly support energy research by providing better constraints on the fundamental physics governing the universe, which can inform models of energy production and consumption on cosmic scales. Additionally, the advanced data analysis techniques developed here could inspire similar approaches in energy data analysis, potentially leading to more efficient and accurate energy systems.
Source: Physical Review D, Volume 109, Issue 6, Article 063509 (2024)
This article is based on research available at arXiv.