MNRAS 468, 3746–3756 (2017) doi:10.1093/mnras/stx638
Advance Access publication 2017 March 16
The first interferometric detections of fast radio bursts
M. Caleb,
1,2,3‹
C. Flynn,
2,3
M. Bailes,
2,3
E. D. Barr,
2,3,4
T. Bateman,
5
S. Bhandari,
2,3
D. Campbell-Wilson,
3,5
W. Farah,
2
A. J. Green,
3,5
R. W. Hunstead,
5
A. Jameson,
2,3
F. Jankowski,
2,3
E. F. Keane,
6
A. Parthasarathy,
2,3
V. Ravi,
2,3,7
P. A. Rosado,
2,8
W. van Straten
2,9
and V. Venkatraman Krishnan
2,3
1
Research School of Astronomy and Astrophysics, Australian National University, ACT 2611, Australia
2
Centre for Astrophysics and Supercomputing, Swinburne University of Technology, PO Box 218, Hawthorn, VIC 3122, Australia
3
ARC Centre of Excellence for All-Sky Astrophysics (CAASTRO), School of Physics, The University of Sydney, NSW 2006, Australia
4
Max-Planck-Institut f
¨
ur Radioastronomie, Auf dem H
¨
ugel 69, D-53121 Bonn, Germany
5
Sydney Institute for Astronomy (SIfA), School of Physics, The University of Sydney, NSW 2006, Australia
6
SKA Organisation, Jodrell Bank Observatory, Cheshire, SK11 9DL, UK
7
Cahill Center for Astronomy and Astrophysics, MC249-17, California Institute of Technology, Pasadena, CA 91125, USA
8
Monash Centre for Astrophysics, School of Physics and Astronomy, Monash University, VIC 3800, Australia
9
Institute for Radio Astronomy and Space Research, Auckland University of Technology, Private Bag 92006, Auckland 1142, New Zealand
Accepted 2017 March 13. Received 2017 March 13; in original form 2016 November 15
ABSTRACT
We present the first interferometric detections of fast radio bursts (FRBs), an enigmatic new
class of astrophysical transient. In a 180-d survey of the Southern sky, we discovered three
FRBs at 843 MHz with the UTMOST array, as a part of commissioning science during a
major ongoing upgrade. The wide field of view of UTMOST (≈9deg
2
)iswellsuitedtoFRB
searches. The primary beam is covered by 352 partially overlapping fan-beams, each of which
is searched for FRBs in real time with pulse widths in the range 0.655–42 ms, and dispersion
measures ≤2000 pc cm
−3
. Detections of FRBs with the UTMOST array place a lower limit
on their distances of ≈10
4
km (limit of the telescope near-field) supporting the case for an
astronomical origin. Repeating FRBs at UTMOST or an FRB detected simultaneously with the
Parkes radio telescope and UTMOST would allow a few arcsec localization, thereby providing
an excellent means of identifying FRB host galaxies, if present. Up to 100 h of followup for
each FRB has been carried out with the UTMOST, with no repeating bursts seen. From the
detected position, we present 3σ error ellipses of 15 arcsec × 8.
◦
4 on the sky for the point of
origin for the FRBs. We estimate an all-sky FRB rate at 843 MHz above a fluence F
lim
of 11 Jy
ms of ∼78 events sky
−1
d
−1
at the 95 per cent confidence level. The measured rate of FRBs at
843 MHz is two times higher than we had expected, scaling from the FRB rate at the Parkes
radio telescope, assuming that FRBs have a flat spectral index and a uniform distribution in
Euclidean space. We examine how this can be explained by FRBs having a steeper spectral
index and/or a flatter logN–logF distribution than expected for a Euclidean Universe.
Key words: instrumentation: interferometers – methods: data analysis – surveys –
intergalactic medium – r adio continuum.
1 INTRODUCTION
Fast radio bursts (FRBs) are a relatively new class of radio transient
that are short, bright and highly dispersed. The pulses are typi-
cally of durations of a few milliseconds, and exhibit a dispersion
sweep characteristic of propagation through a cold diffuse plasma
E-mail: manishacaleb@gmail.com
(Lorimer et al. 2007; Thornton et al. 2013). The dispersion measures
(DMs) of these pulses are significantly higher than the contribution
from the line of sight through the Galactic interstellar medium
(ISM), suggestive of a cosmological origin in which the large DMs
are due to passage through the intergalactic medium (IGM). If
they are at cosmological distances, their inferred intrinsic energies
(>10
31
J) and brightness temperatures (T
b
> 10
33
K) necessitate
a coherent emission mechanism, while the short durations of the
pulses suggest a very compact source of origin (Dennison 2014;
C
2017 The Authors
Published by Oxford University Press on behalf of the Royal Astronomical Society
First interferometric detections of FRBs 3747
Luan & Goldreich 2014). The 18 FRBs published to date [refer to
the FRBCAT repository
3
for the complete list (Petroff et al. 2016)]
have been discovered in either post-processing of archival surveys
or, in real time, using the Parkes radio telescope with the exception
of two, detected at the Arecibo (Spitler et al. 2014) and Green Bank
telescopes (GBT) (Masui et al. 2015). All but one of the bursts have
been found at 1.4 GHz, with the exception being the GBT burst,
which was seen at 800 MHz.
The observed FRB all-sky rate is very high. Champion et al.
(2016) derive a rate of 7
+5
−3
× 10
3
events sky
−1
d
−1
at 1.4 GHz for
bursts between 0.13 and 1.5 Jy ms in fluence and widths in the range
0.128–16 ms. The high FRB rate is a major constraint on theories
for their origin. Until recently, such theories have generally assumed
they are cataclysmic events, in which the progenitor is obliterated.
However, one FRB is now known to repeat in a non-periodic manner
(FRB 121102; Spitler et al. 2016), opening up possibilities for other
progenitor models. Following the discoveries reported in this paper,
Chatterjee et al. (2017) have achieved sub-arsecond localization of
the FRB 121102 using radio interferometric observations from the
Very Large Array. The source has been localized to a m
r
= 25.1AB
mag low-metallicity, star-forming dwarf galaxy at z = 0.192 73(8)
(Tendulkar et al. 2017). The precise localization shows that the
source is either co-located with a 180 µJy active galactic nucleus
or an unresolved type of extragalactic source. However, the exact
nature of the FRB progenitor is still unknown.
Despite concerted follow-up efforts for almost all FRBs, this re-
mains the only FRB seen to repeat. These efforts have been quite
substantial. For instance, ≈80 h of follow up for the Lorimer burst
(Lorimer et al. 2007), ≈80 h for FRB 131104 (Ravi, Shannon &
Jameson 2015)and≈110 h of selected FRB positions (Petroff
et al. 2015) at the Parkes radio telescope yielded no repeats. This
suggests the possibility of there being two independent classes of
FRBs – repeating and non-repeating – with two classes of possible
progenitors (Keane et al. 2016). Progenitor theories include flaring
magnetars (Lyubarsky 2014), giant pulses from pulsars (Cordes &
Wasserman 2016; Connor, Sievers & Pen 2016a; Lyutikov, Burzawa
& Popov 2016), binary white dwarf mergers (Kashiyama, Ioka &
Mesaros 2013), neutron star mergers (Totani 2013) and collapsing
supramassive neutron stars (Falcke & Rezzolla 2014). It is possi-
ble that the lack of repetition of pulses for the FRB discoveries at
the Parkes radio telescope is merely due to limited sensitivity and
follow-up time, and that all FRBs have a common origin (Scholz
et al. 2016). FRB 010724 is an exception to this; however, its ex-
treme brightness (∼30 Jy) far outweighs the lower gain of Parkes
relative to Arecibo, so that one cannot infer its lack of repeat bursts is
due to limited sensitivity. Recently, Ravi et al. (2016) have reported
the detection of FRB 150708, which is of comparable brightness
(∼12 Jy) to FRB 010724, and exhibits 100 per cent polarization
and suggests weak turbulence in the ionized IGM. DeLaunay et al.
(2016) have associated a γ -ray transient with the FRB 131104 dis-
coveredbyRavietal.(2015). However, Shannon & Ravi (2017)
in contrast, report on the discovery of a variable source (consistent
with an AGN) temporally and spatially coincident with the FRB
131104 but not spatially coincident with the γ -ray burst, and rule
out the association of the γ -ray burst with the FRB using proba-
bilistic reasoning.
Most published FRBs have been detected with single dish an-
tennas, with relatively poor angular resolution, and we are unable
to indisputably rule out a near-field or atmospheric origin for the
one-off events until now. The FRB detections made with the multi-
beam receiver at the Parkes radio telescope however, are likely to
originate at 20 km (Vedantham et al. 2016
). Also FRB 150418
has been proposed to be associated with a galaxy at z ∼ 0.5. How-
ever, this association has been called into question by Williams
&Berger(2016) and Vedantham et al. (2016), and other models
like giant pulses from extragalactic pulsars which could account
for the excess DM in the local environment, have been proposed
(Connor et al. 2016a). Better localization during discovery in the
radio requires an interferometric detection.
In a companion paper, we describe how the Molonglo Obser-
vatory Synthesis Telescope (sited near Canberra in Australia) is
currently undergoing a major upgrade, with the addition of a state-
of-the-art correlator to transform it into an FRB finding machine –
the UTMOST (Bailes et al., submitted). Two FRB searches were
performed with UTMOST in 2015 during the upgrade, when the
system was operating at a small fraction of the final expected sen-
sitivity, and only yielded an upper limit of the FRB rate (Caleb
et al. 2016b).
We have now undertaken a third FRB survey at UTMOST and
discovered three FRBs. These are the first FRBs observed with an
interferometer, further strengthening the case for an astronomical
origin in addition to the detections at other telescopes and in the
expected number of beams at Parkes for far-field events, as de-
tection with UTMOST implies the events are in the far-field region
10
4
km. Section 2 of this paper briefly outlines the telescope spec-
ifications, survey properties and the transient detection pipeline. We
present the bursts’ properties and their follow-up observations and
localization areas in Section 3. The event rate estimates of the
FRBs at 843 MHz based on the detections of the three FRBs and
constraints on their spectral index are detailed in Section 4 followed
by our conclusions in Section 5.
2 UTMOST SPECIFICATIONS AND SURVEY
PROPERTIES
The UTMOST consists of an east–west (E–W) aligned cylindrical
paraboloid divided into two ‘arms’ (separated by a 15-m gap), each
11.6-m wide and 778-m long, with 7744 right circularly polarized
ring antennas operating at 843 MHz on a line feed system at its
focus. Groups of 22 consecutive ring antennas (these groups are
termed ‘modules’) are phased to the physical centre of the module,
forming 352 unique inputs (each with a beam 4.
◦
0 × 2.
◦
8FWHP)
that are then beamformed (Bailes at al., submitted). We operate
the telescope by tilting the arms north–south and steering the ring
antennas east–west by differential rotation. UTMOST can access
the sky south of δ =+18
◦
with the east–west steering limited to
±60
◦
. The telescope’s field of view, sensitivity and high duty cycle
make it a near ideal survey instrument for finding FRBs and other
radio transients. Since late 2015, we have been using UTMOST to
search for fast radio transients for an average of 18 h a day, while
simultaneously timing more than 300 pulsars weekly (Bailes et al.,
in preparation, Jankowski et al., in preparation).
In FRB search mode, the 4.
◦
0 FWHP of the primary beam is tiled
in the E–W direction by 352 elliptical, coherent, tied-array beams
(called ‘fan-beams’ or FBs, each 46 arcsec wide), spaced 41 arcsec
apart and overlapping at very close to their half power points at
843 MHz. In the N–S direction, the resolution of the FBs is the
same as that of the primary beam (≈2.
◦
8). The FBs are numbered
from 1 to 352 running from east to west across the primary beam,
with FB 177 directly centred on boresight. The sensitivity of the
telescope to bursts can be estimated using the radiometer equation:
S
min
= β
(S/N
min
) T
sys
G
ν W N
p
(1)
MNRAS 468, 3746–3756 (2017)
3748 M. Caleb et al.
where S
min
is the minimum detectable flux for a threshold signal-
to-noise S/N
min
, β is the digitization factor, ν is the bandwidth
in Hz, N
p
is the number of polarizations (N
p
= 1 for UTMOST as
it is right circularly polarized only), W is the pulse width in ms,
T
sys
is the system temperatures in K and G is the system gain in
KJy
−1
. We define S/N as the ratio of the sum of the on-pulse flux
to the product of the rms of the off-pulse flux and square root of
number of on-pulse bins (S/N =
I
on
√
nbin I
off
). For the fully upgraded
instrument, we expect S
min
= 1.6 Jy ms for a 10σ 1-ms wide pulse,
3.5 K Jy
−1
gain, 100 K system temperature and 31.25 MHz band-
width. The system bandwidth is however only about half of the ini-
tially anticipated 31.25 MHz bandwidth, as the ring antennas have
a significant roll-off in sensitivity away from 843 MHz. This has
been measured using integrated pulses from the pulsar J1644−4559.
Wefindthatonaverage∼86 per cent of the total S/N is concen-
trated in the upper half of the band (∼836–850) as the antennas
are tuned to maximum sensitivity at 843 MHz. We adopt a band-
width of 16 MHz for the sensitivity calculations in the paper, to be
conservative.
During the upgrade, we characterize the system sensitivity by a
fraction of the final expected gain . This factor encompasses sys-
temic losses due to (1) pointing errors (from physical misalignment
in the modules N–S, and phasing errors in the antenna system E–
W), (2) self-generated radio frequency interference (RFI) mainly
due to improperly shielded electronics in the receiver boxes near
the telescope, (3) coherent noise in the receiver boxes, which affects
some sets of adjacent modules, and other inefficiencies in the sys-
tem performance that we are still characterizing, such as systematic
errors in the phase/delay solutions across the interferometer (Bailes
et al., in preparation).
At present (2016 October), we estimate ≈ 0.14, based on obser-
vations of strong calibrators of known flux densities and a number
of high DM pulsars with relatively stable flux densities. This implies
an effective T
sys
of 400 ±100 K. This is significantly higher than the
system temperature seen on the best performing modules, which can
be as low as 100 K. We note that can vary from day to day as mod-
ules are either serviced in the field or have electronics maintenance
in the workshops, and typically lie in the range 0.15 <<0.20.
Occasionally, if only one arm is operational, we have the option
to continue surveys at half sensitivity (i.e. 0.07 <<0.10). The
telescope can access the southern sky for δ<+18
◦
, and for most
parts of the sky we tend to observe reasonably close to the merid-
ian, in order to maximize sensitivity. The sensitivity is reduced by
projection effects away from the meridian.
In 2015 November, we commenced our third FRB survey ‘V3.0’.
It ran for a total of 159.0 d on sky (between 2015-11-01 and 2016-
11-30), at ≈ 0.14 of the final target telescope sensitivity. Our
fluence limit of the survey, that is the fluence of the narrowest
detectable pulse F
lim
can be parametrized as
F
lim
≈ 11
W
ms
1/2
Jy ms (2)
where, 11 Jy is the UTMOST flux limit for S/N = 10, G = 3.0 K
Jy
−1
, ν = 16 MHz, W = 1ms,N
p
= 1andT
sys
= 400 K. It should
be noted that this is not the same as the fluence completeness limit
F
complete
. Between F
lim
and F
complete
, we are incomplete and not all
FRBs with fluences in this range are detectable. This incomplete-
ness region corresponds to the pink shaded region in Fig. 7. The two
previous surveys (V1.0 and V2.0) reported in Caleb et al. (2016b)
yielded no FRB events. Relative to V3.0, V1.0 ran for 19.5 d at
lower sensitivity ( = 0.07), while V2.0 operated for 9.4 d at the
same sensitivity ( = 0.14). FRB survey V3.0 consists primarily
of pointings taken commensally during pulsar timing observations.
In this mode, the time series data from 352 FBs are searched for
dispersed single pulses in real time, using a custom version of the
HEIMDALL software on 8 Nvidia GeForce GTX TITAN X (Maxwell)
GPUs with a latency of 8-s. The resulting candidates were then pro-
cessed offline, typically the following morning for overnight pulsar
timing (RFI is much reduced at night, and the telescope is made
available for maintenance on week days). On weekends, the tele-
scope is usually operated continuously. The candidate processing
pipeline used is described in detail in Caleb et al. (2016b). The
process followed is:
(i) obtain 352 data streams (8-bits/sample), one for each FB, at
655.36-µs sampling;
(ii) search time series for single pulses with width,
0.655 36 < W < 41.943 ms (W = 2
N
× 0.655 36 ms, where
N = 0,1,2,...) and DMs in the range 100 < DM < 2000 pc cm
−3
;
(iii) remove events occurring simultaneously in more than three
FBs at a given instant in time;
(iv) classify only events with S/N ≥10, DM ≥100 pc cm
−3
and
W ≤ 41.943 ms as potential FRB candidates. These then require hu-
man scrutiny of the diagnostic plots, to remove candidates that were
RFI, almost always due to narrow-band mobile handset emissions
in our operating passband and single pulses from known pulsars.
3 RESULTS
The false positive rate at UTMOST is high due to RFI caused by mo-
bile phone handsets, which produce narrow band (5-MHz) emission
in our band, typically in ≈20 ms pulses. These can be eliminated
because celestial pulses are expected to be broad-band, modulated
by a frequency dependent response across the 31.25 MHz band-
width. This process has been validated using individual pulses from
about 20 bright pulsars seen to date. We are presently automating
this process using machine learning algorithms, s o that pulses can
trigger a full voltage dump of the raw data while they are still in
the ≈30 s of ring buffer storage, with alerts issued in near real time.
RFI occurs predominantly at low DM, but the rate is high enough to
produce a few hundred spurious candidates above our DM limit of
100 pc cm
−3
daily. Candidates were typically vetted each morning
after data taking.
In 2016 March, April and June, we made the first interferometric
detections of FRBs at 843 MHz: FRB 160317, FRB 160410 and
FRB 160608, as shown in Fig. 1.
3.1 FRB 160317
This was detected on 2016 March 17 at 09:00:36.530 UTC while
observing an X-ray magnetar SGR 0755−23, in response to an
Astronomers Telegram (Barthelmy et al. 2016). The burst occurred
about 0.
◦
4 east of the magnetar, and was detected ∼1
◦
off the Galactic
plane with a DM of 1165(11) pc cm
−3
. The DM due to the ISM at
this sight line is ∼320 pc cm
−3
from the NE2001 model by Cordes
& Lazio (2002)and∼395 pc cm
−3
from the YMW16 model (Yao,
Manchester & Wang 2017). The burst with S/N ∼13, occurred east
of the centre of the primary FB of detection (Beam 212) since it
appeared weakly in the adjacent FB with S/N ∼5 (Beam 213) as
shown in Fig. 2.
MNRAS 468, 3746–3756 (2017)
First interferometric detections of FRBs 3749
Figure 1. Frequency versus time behaviour of FRBs 160317, 160410 and
160608 detected at UTMOST at the centre frequency of 834.765 MHz.
The top panel in each case shows the frequency-averaged pulse profile. The
bottom panel shows that narrow-band RFI has been excised and the effects
of interchannel dispersion have been removed assuming DMs of 1165 ±11,
278 ± 3 and 682 ± 7pccm
−3
, respectively. The data are uncalibrated as
the bandpass of the system varies as a function of the meridian angle, and
the flux densities are in arbitrary units. Note the different time range on the
abscissa for FRB 160410.
Figure 2. The three panels display the total power pulse profiles for one
polarization in three adjacent FBs. FRBs 160317 and 160410 were also
detected as sub-threshold events in neighbouring FBs (in addition to the
high S/Ns in the primary detection FBs), indicating that they did not occur
near the centres of the primary FB. On the contrary, FRB 160608 was only
detected in one FB suggesting that it occurred close to the centre of beam
208 (see bottom panel).
MNRAS 468, 3746–3756 (2017)
3750 M. Caleb et al.
Figure 3. The sky distribution of the 18 FRBs published to date in Galactic coordinates. Dots mark the positions of the FRBs detected at the Parkes telescope,
the triangle represents FRB 121102 detected at the Arecibo telescope and the square represents FRB 110523 discovered at the GBT. Stars mark the positions
of the UTMOST FRBs. Two of the Parkes FRBs have positions separated by 9 arcmin which are not resolved in this figure. It should be noted that there are
large biases in this distribution due to very different sky coverages and survey depths.
3.2 FRB 160410
Similar to FRB 160317, this FRB was also detected in two adjacent
FBs (Beam 085 with S/N ∼13 and Beam 084 with S/N ∼4) as seen
in Fig. 2. A single dispersed pulse was discovered on 2016 April 04
at 08:33:39.680 UTC, in an observation of the pulsar J0837+0410
at the telescope’s boresight. This pulsar is so bright that individual
pulses were seen from it as the FRB occurred, meaning the flux
density scale and bandpass response of the observation were well
understood. The FRB was seen ∼1
◦
away from boresight. This pulse
was detected at Galactic latitude, ∼27
◦
with the line-of-sight DM
accounting for only ∼58 pc cm
−3
of the total observed DM from
the NE2001 model. The YMW16 model estimates ∼63 pc cm
−3
.
FRB 160410 has one of the lowest DM excess’ ∼220 pc cm
−3
till
date making it one of the closest known FRBs and an excellent
candidate to search for repeat pulses.
3.3 FRB 160608
The burst occurred in an observation of the pulsar J0738−4042 at
l = 254.
◦
11 and b =−9.
◦
54 on 2016 June 06 at 03:53:01.088 UT
with a total DM of ∼682 pc cm
−3
and ∼238 pc cm
−3
contribution
from the Milky Way (NE2001). The YMW16 model’s estimate
however is ∼310 pc cm
−3
. It was seen ∼0.
◦
5 from the boresight
position. FRB 160608 was detected with S/N ∼12, just above the
detection threshold of 10 and it occurred towards the centre of the
primary detection FB (Beam 208). No pulse was detected in the
adjacent FBs (see Fig. 2). This was initially of concern, but tests
with the Vela pulsar placed sufficiently far south of the telescope
boresight, to produce an individual pulse with the same S/N showed
that detection in a single FB occurred ≈20 per cent of the time. The
localization of this FRB is thus slightly poorer (21 arcsec × 8.
◦
4)
than for the other two FRBs, for which a two FB detection allow a
more accurate position.
The sky distribution of the 3 FRBs in Galactic coordinates, with
respect to the positions of other published FRBs is shown in Fig. 3.
All 3 FRBs have been localized to narrow ellipses on the sky with
their orientation hour angle dependent as seen in Fig. 4. The primary
advantage of the array is that a pulse from a far-field point source
is detected in a maximum of three adjacent FBs at any given time,
confirmed by extensive pulsar observations. RFI is typically near-
field, and predominantly appears in more than three adjacent FBs,
meaning that it can be reliably excised to reduce false positive rates
when searching for transients. Using the adjacent FB detections of
FRB 160317, we have modelled the point of separation between the
near-field or Fresnel region and the far-field or Fraunhofer region
of the telescope. Assuming a point source at 10
6
km, we compute
the S/N for a tied-array beam (e.g. FB 212) phased at an offset of
0.3 from the centre of the beam to ensure a two FB detection. We
compute the path length to each module, the phase of the signal
along the array and perpendicular to the array, and add all these
as a vector sum weighted by the module performance, to get the
‘boresight’ S/N. We see that in Fig. 5 at a distance of 10
4
km, we
achieve a two FB detection with S/N ∼13 in the primary detection
beam and S/N ∼5 in the secondary detection beam, similar to the
FRB being modelled. Detections of FRBs in one or two FBs only,
thus allows us to identify them as sources more distant than this,
placing them well away from the Earth and hence effectively rule
out sources of local origin.
The discovery observations containing the FRBs were care-
fully inspected to check for similar events at the same time and
with the same DM as the FRB, in other FBs. No other broad-band
pulses were detected in any other FBs within approximately 60 s
of the bursts. Moreover, in addition to all the tied array FBs, we
form a single special FB as the incoherent sum of all the other
FBs. This ‘total power’ FB was also searched for events near
the UTC of the three bursts. For the three FRBs, this FB con-
tained no unusual sources of RFI. Only twice during the three
surveys did we find FRB-like candidates (i.e. appearing across the
band) that were identified as RFI upon closer analysis. In each
case, similar events could be found in dozens to hundreds of FBs,
and were thus obvious near-field RFI. These false candidates also
had ‘patchy’ power across the observing band, indicative of RFI
MNRAS 468, 3746–3756 (2017)
