First SETI Observations with China's Five-hundred-meter Aperture Spherical Radio Telescope (FAST)

Brief

In July 2019, the UC Berkeley SETI group used FAST's 19-beam L-band receiver commensally with the SERENDIP VI 128-million-channel spectrometer (1000–1500 MHz, ~3.725 Hz resolution) to collect five hours of drift-scan commissioning data at 38 GB/s raw throughput. The Nebula offline pipeline applied zone, drifting, and multi-beam RFI classifiers followed by k-nearest-neighbor (KNN) and DBSCAN clustering to clean the data and select candidates. Injecting 294 synthetic 'birdie' signals across 20 groups, the pipeline removed 99.9063% of all hits while losing only 5.1020% of birdies, and DBSCAN recovered 277 of 294 birdies (94.2177%). Eighty-three additional candidate groups were identified; two near 1055 MHz showed no surrounding hits across five hours and are flagged as the most interesting, though the paper makes no ETI claim.

Metadata

Category

Search

Venue

preprint (arXiv astro-ph)

Type

Preprint

Year

2020

Authors

Zhi-Song Zhang, Dan Werthimer, Tong-Jie Zhang, Jeff Cobb, Eric Korpela

arXiv

2002.02130

Access

Open access

Length

6.8 M

Programs

SERENDIP VI, Nebula, CRAFTS, Breakthrough Listen, SETI@home

Instruments

FAST 19-beam L-band receiver, SERENDIP VI spectrometer, CASPER ROACH2 FPGAs

Data sources

FAST commissioning drift-scan survey (July 2019), synthetic birdie injections (294 signals, 20 groups)

Tags

SETI, technosignature, radio astronomy, RFI mitigation, machine learning, narrowband search, commensal survey

Key points

• FAST's effective sensitivity is ~1800 m²/K versus Arecibo's ~1100 m²/K, a ~64% advantage, and its declination coverage (−14.3° to +65.7°) substantially exceeds Arecibo's (+1.5° to +38.5°).p.2
• SERENDIP VI provides 128 million spectral channels at ~3.725 Hz resolution across 1000–1500 MHz; any channel exceeding S/N > 30 is recorded as a 'hit' with time, frequency, power, beam, and pointing metadata.p.3
• The 19-beam receiver at 1 Gsps sampling produces 38 GB/s of raw voltage data, making offline bulk storage infeasible and requiring GPU-based real-time FFT reduction across 19 compute nodes.p.2
• The combined Nebula + KNN pipeline removes 99.9063% of all recorded hits; zone RFI (stable narrow-band terrestrial interference) accounts for 98.1976% of the flagged total, with civil aviation occupying 960–1215 MHz as the dominant source.p.12
• Birdie-injection validation: DBSCAN recovers 277 of 294 synthetic signals (94.2177%), losing only one complete group to Nebula; the KNN 90% threshold was calibrated to maximize birdie retention while suppressing residual RFI clusters.p.13
• Two candidate groups near 1055 MHz. Group 1 (5 hits, single channel, ~5 s, beam 15) and Group 2 (80 hits, six successive channels, ~18.6 Hz bandwidth, ~20 s, beams 14–15), have no surrounding hits in the full five-hour dataset.p.13
• Multi-beam RFI rejection flags signals appearing simultaneously in non-adjacent beams: a genuine point-source would illuminate at most one or two adjacent beams, while terrestrial RFI typically floods multiple non-adjacent beams.p.5
• The planned CRAFTS commensal survey will cover 57% of the celestial sphere over more than 5,000 telescope hours, with the SETI pipeline running continuously on copied data streams.p.2

Verbatim

• “The probability of success is difficult to estimate, but if we never search the chance of success is zero”

  p.1

• “FAST's sensitivity is ∼ 1800 m 2 /K (Arecibo is about 1100 m 2 /K )”

  p.2

• “We are very happy to note that our pipeline found 277 birdies, 94.2177% of the total.”

  p.13

Most interesting

• The 19-beam receiver generates 38 GB/s of raw data, over 130 TB per hour, making any storage-first approach physically impossible and requiring the entire RFI classification chain to run in real time.
• Zone RFI constitutes 98.1976% of all flagged data; civil aviation alone occupies a 255 MHz block (960–1215 MHz) that overlaps roughly half of FAST's L-band SETI search window.
• Astrophysical masers (~500 Hz minimum width) set the observational floor for 'natural' signal bandwidth; any signal narrower than this is presumed artificial, the operational definition underpinning every narrowband SETI search.
• Real-time telescope pointing was unavailable during the July 2019 commissioning run; reconstructed positions showed total deviation of ~50 arc seconds over five hours, or 0.0028 arc seconds per second on average.
• The two most anomalous candidate groups both cluster near 1055 MHz, a frequency absent from FAST's catalogued RFI source list, and are separated by approximately 150 seconds in time with no intervening hits.
• The FAST SETI instrument was physically installed by the UC Berkeley SETI group in September 2018, making it the first operational foreign-led commensal back-end at the facility.