Canada's new radio telescope, built to explore the early universe, has turned out to be a handy hunter for the mysterious phenomenon called Fast Radio Bursts (FRB).
These bursts from outer space were first picked up in 2007 (in data archived since 2001) by Australia's Parkes radio telescope, and only a few dozen have been detected since then.
Now the Canadian Hydrogen Intensity Mapping Experiment (CHIME) has set the astronomy world afire in announcing its findings in two papers in Nature, that during its pre-commissioning phase in July and August 2018, it spotted 13 new FRBs.
That's a decent first-pass contribution to a phenomenon that's been spotted so rarely. Perimeter Institute astronomer Dr Dustin Lang said on Wednesday this week: “We have more ideas of what they could be than we have actually detected fast radio bursts.”
Two of the new observations are particularly special: one, by showing the lowest “dispersion measure” ever reported for an FRB, is probably among the closest observed and the other is just the second repeating FRB ever identified.
In a message posted to Twitter, astrophysicist Emily Petroff, who in 2015 was the first to identify an FRB in real time, noted that the new “repeater” behaves a lot like the first:
But possibly most exciting of all... The repeater found by CHIME shows similar frequency and time structure to the only other repeater. The emission cascades down in frequency over time. Figures from CHIME paper 2 (left) and Hessels et al. 2019 (right) https://t.co/hfWhDl2FyB pic.twitter.com/ecL817SHKH— Emily Petroff (@ebpetroff) January 9, 2019
The search for FRBs has a simple motivation: so far, we have too few observations to settle on a cause for the phenomenon. An FRB is a very short pulse, akin to the kind of short pulses that come from pulsars, but unlike pulsars they mostly don't repeat.
They're probably generated by extremely energetic events, since most FRBs are detected from a great distance. Seeing a signal at all indicates something big, like a black hole collision, could be the cause.
CHIME is useful for the FRB search as it's a static instrument that relies on the Earth's rotation to sweep its view across the sky and this allowed its designers to build an instrument much larger than can easily be achieved with a swiveling dish design.
Taming the firehose
CHIME collaborator and Perimeter Institute fellow Kendrick Smith explained that using the telescope to search for FRBs poses a huge computational challenge.
In a Perimeter Institute video (below), Smith said the telescope generates an “avalanche of data, a hundred times more data than is generated by any other radio telescope.” To process that, the team had to work out “how to scale the computations that are needed to do radio astronomy up to unprecedented volume.”
The processing pipeline behind CHIME is described in this paper on arXiv.
One reason CHIME was set on this search was that since the world first started detecting the phenomenon, astroboffins have upped their estimates of how many such events exist to somewhere between hundreds and thousands a day.
Only, however, if we can see them, and that's a multifaceted problem: First, deploy a detector able to pick up the signals and second, pluck out the transients from a torrent of noise (CHIME collects a bonkers 1TB of data per second for FRB detection. The 1,024 stationary intensity beams use 16,000 frequency channels, sampled each millisecond).
Turning that incoming firehose to the mere 142GB per second to the back end is a computational problem, and this is why the Perimeter Institute's fellow Kendrick Smith called CHIME “a software telescope”.
After pre-processing, discrete beams are processed by 128 “Layer 1” nodes, each comprising two ten-core Intel Xeon E5-2630 v4 processors and 128GB of RAM. In this stage, RF interference is rejected, a signal-correction process called dedispersion is applied, and possible signals are sifted and grouped.
The paper noted that this dedispersion transform, in which signals are converted from time and frequency, into time and dispersion measure, to allow “efficient detection of dispersed impulse signals”.
The CHIME arXiv paper explained that “the output of the dedispersion transform is a 5D array of signal-to-noise ratios (SNRs), to which we apply a tunable threshold … to identify candidate events for processing by subsequent stages of the pipeline”.
Take a deep breath, because here come some breathtaking numbers from the CHIME arXiv paper: “The computational challenge of the CHIME/FRB search is immense: The input data rate is 1.5PB/day, and the dedispersion transform computes 1011 SNR values per second (total for all beams).”
To handle this processing stage, the CHIME team developed its own software, dubbed Bonsai, which it plans to release in the future.
The remaining steps in the CHIME pipeline are L3 (flux estimation, source identification, extragalactic check, and an action decision); and L4 (action implementation, database operations to store header data, intensity data, and baseband data; offline analysis and a Web interface with alerts). ®