A spatial filter for mitigating radio interference and its application to CHIME/FRB Outriggers

Abstract

The sensitivity of radio telescopes is becoming increasingly limited by the presence of radio frequency interference (RFI), which will worsen as the radio spectrum becomes more crowded. One context where this poses a challenge is the field of fast radio burst (FRB) science, where there is increasing scientific interest in capturing as large of a population of bursts as possible and accurately measuring their celestial coordinates using interferometry. With several modern radio facilities actively collecting data for large FRB surveys that will be transformative to the field, properly mitigating unwanted interference is essential for the science goals of these surveys to be met. In this work, we present variations of a spatial filter based on the Karhunen-Loeve (KL) Transform to enhance the sensitivity of radio interferometers and demonstrate its applicability to FRB detection and localization. We derive a particular variation of the filter for the case of point-like radio pulses, which we show reduces to the maximum-signal-to-noise beamformer. We apply this filter to CHIME/FRB baseband data and demonstrate its capability to enhance the sensitivity and overall localization rate of CHIME/FRB Outriggers. We compare the cross-correlation signal-to-noise obtained using the spatial filter with that obtained using a spectral-kurtosis RFI flagger for a sample of 100 FRBs recorded by CHIME and its Outriggers, and show that this filter will double the total number of FRBs successfully localized with the CHIME/FRB Outrigger telescopes. While demonstrated here in the context of CHIME/FRB Outriggers, the spatial filter presented in this work--which we have made publicly available--is broadly applicable to other interferometric radio facilities engaged in FRB science and transient detection, including next-generation telescopes such as CHORD, DSA-2000, BURSTT, and CHARTS.

Figures (10)

• Figure 1: Top panel: comparison of a beamformed pulse from FRB 20210603A before (left) and after (right) applying the KL filter for the top 100 MHz of the CHIME band. The RFI contaminated channels in this part of the band are highlighted in red. This demonstrate the filter's ability to recover a pulse in channels strongly contaminated by RFI. Bottom panel: here we denote the inverse of the digital gains---averaged over all receiver and digitizer chains---as the "digital bandpass correction", which are shown in units of dB as a function of frequency. RFI contaminated channels, which are shaded red, are clearly visible from the significant jumps in the digital gains, where the average RFI power levels are over 10 dB larger than the noise in clean channels. In dark red are the RFI contaminated channels spanning $\sim730-756$ MHz.
• Figure 2: Top panel: comparison of a beamformed pulse from FRB 20210603A before (left) and after (right) applying the KL filter over the full band. Middle and bottom panels: Improvement in the signal-to-noise (ratio) of the beamformed pulse (in units of dex) after applying the filter as a function of frequency. Some RFI-contaminated channels (identified in the bottom panel of Figure \ref{['fig:morphology']}) are shaded in red, where we see the most notable signal-to-noise improvement.
• Figure 3: Comparison of the non-overlapping RFI environments at HCO, GBO and CHIME, represented by the inverse of the digital gains averaged over all receiver and digitizer chains. At each individual station, $\sim$ 20-30% of the band on average is lost due to persistent RFI, and about half of the band is lost to RFI in the CHIME-Outriggers cross-correlated data.
• Figure 4: Comparison of cross-correlation signal-to-noise in CHIME-Outrigger cross-correlated VLBI data of pulsar B2154+40, calibrated using J2154+4515. The correct solution corresponding to the known position of the pulsar Chatterjee_2009 is at 0 ns, which is clearly inconsistent with the rfi-contaminated fringes in red.
• Figure 5: Comparison of phase residuals in CHIME-HCO cross-correlated VLBI data for FRB 20250316A RBFLOAT_2025ApJ without the KL filter (top) and with the KL filter (bottom). Note that in both panels, only frequencies not recorded by CHIME itself have been masked out. The residual $\sim$ 30Mhz ripple is due to a differential beam phase kko_adamOutriggers_2025 and is not an artifact of RFI.