Enhanced Bayesian RFI Mitigation and Transient Flagging Using Likelihood Reweighting

TL;DR

This paper introduces a Bayesian approach with likelihood reweighting for efficient, automated mitigation and detection of transient RFI in radio astronomy data, significantly reducing computation time.

Contribution

It presents a novel likelihood reweighting technique that makes Bayesian transient RFI mitigation computationally efficient and automates flagging threshold fitting.

Findings

25-fold reduction in computation time for large data sets

Automated flagging threshold fitting prevents overcorrection

Reliable detection of high SNR transient anomalies

Abstract

Contamination by Radio Frequency Interference (RFI) is a ubiquitous challenge for radio astronomy. In particular, transient RFI is difficult to detect and avoid, especially in large data sets with many time bins. In this work, we present a Bayesian methodology for time-dependent, transient anomaly mitigation. In general, the computation time for correcting for transient anomalies in time-separated data sets grows proportionally with the number of time bins. We demonstrate that utilising likelihood reweighting can allow our Bayesian anomaly mitigation method to be performed with a computation time close to independent of the number of time bins. In particular, we identify a factor of 25 improvement in computation time for a test case with 2000 time bins. We also demonstrate how this method enables the flagging threshold to be fit for as a free parameter, fully automating the mitigation process. We find that this threshold fitting also prevents overcorrecting of the data in the case of wide priors. Finally, we investigate the potential of the methodology as a transient detector. We demonstrate that the method is able to reliably flag an individual anomalous data point out of 302,000 provided the SNR > 10.