Breakthrough Listen: Machine Learning Enables New Detections of FRB 121102

Data and Technical Tutorial

The following data files are spectrograms, intensity or total power (Stokes I) as function of frequency and time, for the 10 scans analyzed in our latest work. The total duration of time covered by these observations is around 6 hours. The data are stored in filterbank format and can be read with the Breakthrough Listen I/O Methods for Python (blimpy) package, as well as other software. These data products have 350 ms and 360 kHz resolutions in time and frequency. An inference script with an example trained CNN can be found on GitHub.

Note these files are large. Smaller snapshots of individual bright pulses are available here.

A list of all individual bursts detected during these scans, including those found by conventional methods and the CNN used in our most recent work, (Table 2 from Zhang et al.) is available in .csv format.

Scan Number    MJD Start         Download Link    Size
11 57991.49905 Download 76G
12 57991.51723 Download 76G
13 57991.53535 Download 76G
14 57991.55354 Download 76G
15 57991.57166 Download 76G
16 57991.58976 Download 76G
17 57991.60787 Download 76G
18 57991.62598 Download 76G
19 57991.64409 Download 76G
20 57991.66219 Download 53G


If you make use of these data for a publication, please consider citing the following references:

[1] Gajjar, Vishal; Siemion, Andrew; MacMahon, David et al. 2018, AAS, 231
[2] Gajjar, Vishal; Siemion, Andrew; MacMahon, David et al. 2017, ATel, 10675
[3] Gajjar, Vishal; Siemion, Andrew; Price, Daniel, et al. 2018, ApJ, 863, 2
[4] Zhang, Yunfan Gerry, Gajjar, Vishal; Foster, Griffin et al. 2018, ApJ Accepted

Datasets are released under the CC BY 4.0 license.

Background Art by Danielle Futselaar