Radio frequency interference cleaning techniques for Meerkat Hi intensity mapping data

Abstract

The detection of the redshifted 21 cm Neutral Hydrogen (HI) emission line is a primary scientific goal of the MeerKLASS survey; however, the intrinsic faintness of this cosmological signal makes it highly susceptible to Radio Frequency Interference (RFI). While standard amplitude-based flagging algorithms, such as SumThreshold, effectively mitigate bright, high-power interference, they often fail to detect sub-thermal, low-level contaminants. These undetected residuals introduce additive biases in the HI auto-power spectrum, hindering the high-precision calibration required to measure Baryon Acoustic Oscillations (BAO). This thesis investigates the adaptation and application of the Sky-Subtracted Incoherent Noise Spectrum (SSINS) algorithm, originally designed for interferometric data, to MeerKAT single-dish auto-correlation data. By implementing specific sky-subtraction techniques and utilizing antenna incoherent averaging to exploit the Central Limit Theorem, the algorithm effectively reduces the thermal noise floor to expose faint, non-thermal RFI signatures. The method is applied in two distinct rounds: a primary round on raw data to establish a conservative baseline, and a secondary round on pipeline-flagged data to target unflagged residual RFI. We validate the performance of the algorithm using a brightness distribution (logN–logS) analysis, which statistically decomposes the data into Gaussian thermal noise and power-law RFI components. The results demonstrate that SSINS effectively identifies specific low-level RFI morphologies that evade the standard KATcali pipeline, including faint GSM uplink signals and extended broadband emission from satellite transmissions. This work establishes that statistical RFI mitigation is a necessary complement to standard thresholding pipelines, offering a robust pathway to cleaning HI intensity mapping data to the sensitivity levels required for cosmology.