Edge-Thresholding: A New Thresholding Based Real-Time RFI-Mitigation Algorithm for Transient Detection

Abstract

Unraveling the mystery of fast radio bursts (FRBs) - extremely high-energy events occurring outside the Milky Way that last fractions of a second - is a hot topic topic in radio astronomy. Detecting new FRBs is increasingly complicated by RFI as the radio bandwidth becomes more congested and radio telescopes become more sensitive. In this paper we propose, implement and validate a novel RFI-mitigation algorithm designed for FRB detection called edge-thresholding.Modern radio telescopes produce an immense amount of data that needs to be processed in real-time. This requirement invalidates many RFI-mitigation algorithms based upon techniques such as kurtosis, surface-fitting or Fourier transforms due to their computational complexity. This had led to the development of thresholding methods which are computational simple, well suited to GPU acceleration and provide accurate RFI mitigation.In this paper we propose edge-thresholding, the first thresholding based RFI-mitigation algorithm specifically designed for FRB detection. Edge-thresholding works by flagging data within a window if the points minimum difference to a boundary point is above a computed threshold. Linear time-complexity means it scales to the increasing data rates of radio telescopes.Edge-thresholding takes advantage of two features that FRBs exhibit but RFI does not. First, FRBs are generally wider than deleterious RFI and second FRBs have a pseudo Gaussian profile due to the finite frequency resolution of the telescope causing DM smearing.We combine edge-thresholding with a time-domain sigma cut in a unified pipeline. A GPU accelerated implementation of this pipeline is shown to run in the real-time on WSRT data. Additionally, less than 36.86 additional kilobytes of memory are required for 804 megabytes of input data.Using data from WSRT containing simulated FRBs we shown that our pipeline reduces false-positives and increases the number of detected FRBs. These results are further validated by experiments using data from the Crab-Pulsar. These results show that our pipeline could outperform AO-Flagger, the RFI-mitigation pipeline used for WSRT and LOFAR.