2019MNRAS.484.2866O

We investigate the implications of interference detection for experiments that are pursuing a detection of the redshifted 21-cm signals from the epoch of reionization (EoR). Interference detection causes samples to be sporadically flagged and rejected. As a necessity to reduce the data volume, flagged samples are typically (implicitly) interpolated during time or frequency averaging or uv-gridding. This so-far unexplored systematic biases of the 21-cm power spectrum, and it is important to understand this bias for current 21-cm experiments as well as the upcoming Square Kilometre Array EoR experiment. We analyse simulated data using power spectrum analysis and Gaussian process regression. We find that the combination of flagging and averaging causes tiny spectral fluctuations, resulting in 'flagging excess power'. This excess power does not substantially average down overtime and, without extra mitigation techniques, can exceed the power of realistic models of the 21-cm reionization signals in Low-Frequency Array (LOFAR) observations. We mitigate the bias by (i) implementing a novel way to average data using a Gaussian-weighted interpolation scheme, (ii) using unitary instead of inverse variance weighting of visibilities, and (iii) using low-resolution forward modelling of the data. After these modifications, which have been integrated in the LOFAR EoR processing pipeline, the excess power reduces by approximately three orders of magnitude, and is no longer preventing a detection of the 21-cm signals.