| |
 |
| Paper: |
Enhancing the SPANDAK Pipeline: Multibeam RFI Mitigation and Integration of Machine Learning for FRB Detection |
| Volume: |
540, Compendium of Undergraduate Research in Astronomy and Space Science |
| Page: |
38 |
| Authors: |
Derrick Cardenas; Vishal Gajjar |
| DOI: |
10.26624/ELBB3581 |
| Abstract: |
Fast radio bursts (FRBs) are brief, often millisecond-duration radio transients of extragalactic origin whose progenitors remain unknown. Their unpredictable
appearance and high dispersion measures make them striking beacons and powerful
probes of the distant universe. However, their physical nature and sporadic signals
pose substantial challenges to automated detection. While the SPANDAK pipeline used
at the Allen Telescope Array (ATA) has been effective in identifying candidates from
single-beam observations, its previous configuration was unable to ingest multibeam
data, limiting its capacity to automatically reject terrestrial radio frequency interference
(RFI). To address this limitation, we developed an extended version of the pipeline
capable of processing multibeam data, allowing simultaneous use of a primary beam
targeted at a candidate source and adjacent off-beams to identify and suppress spurious signals. This configuration leverages the intrinsic spatial nature of FRBs, which
should appear only in the on-target beam. Our implementation restructured the existing
SPANDAK framework to merge off-beam plots with primary beam detections, enabling
faster processing and reducing the candidate pool. We validated this pipeline using simulated FRB injections embedded in noisy, RFI-dense backgrounds and further tested
the pipeline on real multibeam observations of cataloged pulsar sources taken with
the ATA. Results demonstrate a 52% improvement in processing efficiency for larger
datasets while providing enhanced RFI discrimination through multibeam filtering that
removes thousands of additional false positive candidates per observation. The integration of machine learning classification further reduces false positives by 51−81%, with
the combined multibeam and ML approach achieving up to 15× reduction in candidate
volume compared to single-beam processing. The updated system enhances the array’s
multibeam capabilities, providing a more efficient and scalable approach to FRB discovery. By enabling the direct visualization of RFI across beams, the pipeline also generates labeled training data for future machine learning model improvements, laying the
groundwork for fully automated FRB classification and real-time discovery pipelines. |
|
|
|
|
|
