Speaker
Description
Moving sources of radio-frequency interference (RFI), such as airplanes and satellites, can contaminate interferometric data in ways that are difficult to identify and mitigate with standard flagging approaches. We introduce a new modeling scheme for moving sources of radio frequency interference based on the characteristic imprint they leave in the $uv$-plane. Given a known source trajectory in image space, we show that the corresponding sinc-like structure traced in $uv$-space can be solved for analytically and simulated efficiently. Such fast and accurate RFI models are essential for ultimately subtracting these contaminants from interferometric data directly rather than simply flagging and discarding affected visibilities, thereby preserving a higher fraction of usable data.
We then investigate how the structure of moving-source contamination affects baseline-dependent flagging. Because the sinc-like $uv$-plane response suppresses the RFI signal on some $uv$-modes, per-baseline flagging algorithms may struggle to identify contamination on some baselines. We find that the maximum attenuation between the most and least contaminated baselines only approaches three orders of magnitude, suggesting that sufficiently faint partially flagged moving RFI could leave residual contamination capable of biasing an EoR power spectrum analysis.
As a first case study, we simulate a 12-second, 10 kJy aircraft-reflection event using pyuvsim and inject it into three MWA Phase II EoR0 datasets containing 10, 3.3, and 0.3 hours of otherwise clean data. For each dataset, we compare four different flagging cases: no additional flagging, a manual flag mask targeting the exact contaminated times and frequencies on all baselines, an extended flag mask that flags all frequencies during contaminated times on all baselines, and a baseline-dependent mask produced by AOFlagger. Power spectra are then generated using the FHD/eppsilon pipeline and compared against power specra from the corresponding clean datasets without injected RFI.
We find that all flagging strategies introduce some level of power spectrum bias relative to the clean-data case, even when the RFI is accurately identified. This bias is strongest in the 0.3-hour dataset, suggesting that reduced $uv$-coverage plays an important role in the observed flagging bias. However, comparisons with the unflagged case show that any flagging strategy is still preferable to allowing unmodeled RFI to remain in the data.
Direct comparisons between flagging strategies show that the extended mask produces the smallest power spectrum bias, consistent with Wilensky et al. (2022). In contrast, the manual and AOFlagger masks perform similarly for the event considered here, likely because the 10 kJy source is bright enough to be captured effectively by AOFlagger. While this simulation does not provide a stringent test of the faint-residual-RFI scenario suggested by the analytic attenuation estimate, it demonstrates that the modeling framework developed here can be used to quantify the power spectrum impact of moving RFI and compare mitigation strategies. Future work can extend this approach to a larger amount of fainter events spanning a range of brightnesses.