HI intensity mapping with MeerKAT: 1/f noise analysis

TL;DR

This study analyzes the 1/f noise in MeerKAT radio telescope data, demonstrating that noise can be significantly reduced using SVD, which enhances the detectability of the HI signal in intensity mapping experiments.

Contribution

The paper provides a detailed characterization of 1/f noise in MeerKAT and shows that SVD can effectively mitigate correlated noise, improving HI signal detection.

Findings

1/f noise can be reduced by SVD mode subtraction.

Knee frequency drops from >0.1 Hz to ~3×10^{-3} Hz with 2-mode subtraction.

1/f variations are confined to small regions in time-frequency space.

Abstract

The nature of the time correlated noise component (the 1/f noise) of single dish radio telescopes is critical to the detectability of the HI signal in intensity mapping experiments. In this paper, we present the 1/f noise properties of the MeerKAT receiver system using South Celestial Pole (SCP) tracking data. We estimate both the temporal power spectrum density and the 2D power spectrum density for each of the antennas and polarizations. We apply Singular Value Decomposition (SVD) to the dataset and show that, by removing the strongest components, the 1/f noise can be drastically reduced, indicating that it is highly correlated in frequency. Without SVD mode subtraction, the knee frequency over a $20\,$MHz integration is higher than $0.1\,\rm Hz$; with just $2$~mode subtraction, the knee frequency is reduced to $\sim 3\times 10^{-3}\,{\rm Hz}$, indicating that the system induced 1/f-type variations are well under the thermal noise fluctuations over a few hundred seconds time scales. The 2D power spectrum shows that the 1/f-type variations are restricted to a small region in the time-frequency space, either with long wavelength correlations in frequency or in time. This gives a wide range of cosmological scales where the 21cm signal can be measured without further need to calibrate the gain time fluctuations. Finally, we demonstrate that a simple power spectrum parameterization is sufficient to describe the data and provide fitting parameters for both the 1D and 2D power spectrum.