A new approach to mitigation of radio frequency interference in interferometric data

TL;DR

This paper introduces RfiX, a novel software-based method for mitigating radio frequency interference in interferometric data, enabling the recovery of true signals without high-speed hardware or real-time processing.

Contribution

The paper presents RfiX, a new software technique that identifies and removes RFI patterns from interferometric data, applicable to any interferometer and effective on integrated data.

Findings

Successfully applied to GMRT data at 240 MHz with high image quality

Works without requiring RFI-free data segments

Effective on data integrated over 1-10 seconds

Abstract

Radio frequency interference (RFI) is the principal factor limiting the sensitivities of radio telescopes, particularly at frequencies below 1 GHz. I present a conceptually new approach to mitigation of RFI in interferometric data. This has been used to develop a software tool (RfiX) to remove RFI from observations using the Giant Metrewave Radio Telescope, India. However, the concept can be used to excise RFI in any interferometer. Briefly, the fringe-stopped correlator output of an interferometer baseline oscillates with the fringe-stop period in the presence of RFI. RfiX works by identifying such a pattern and subtracting it from the data. It is perhaps the only purely software technique which can salvage the true visibility value from RFI-corrupted data. It neither requires high-speed hardware nor real-time processing and works best on normal correlator output integrated for 1-10s. It complements other mitigation schemes with its different approach and the regime it addresses. Its ability to work with data integrated over many seconds gives it an advantage while excising weak, persistent RFI unlike most other techniques which use high-speed sampling to localise RFI in time-frequency plane. RfiX is also different in that it does not require RFI-free data to identify corrupted sections. Some results from the application of RfiX is presented including an image at 240 MHz with a Peak/noise ratio of 43000, the highest till date at wavelengths >1m.