The Allen Telescope Array Fly's Eye Survey for Fast Radio Transients
Andrew P.V. Siemion, Geoffrey C. Bower, Griffin Foster, Peter L. McMahon, Mark I. Wagner, Dan Werthimer, Don Backer, Jim Cordes, Joeri van Leeuwen
TThe Allen Telescope Array Fly’s Eye Survey for Fast Radio Transients
Andrew P.V. Siemion , Geoffrey C. Bower , Griffin Foster , Peter L. McMahon , Mark I.Wagner , Dan Werthimer , Don Backer , Jim Cordes , Joeri van Leeuwen [email protected] Abstract
The relatively unexplored fast radio transient parameter space is known to be home to a va-riety of interesting sources, including pulsars, pulsar giant pulses and non-thermal emission fromplanetary magnetospheres. In addition, a variety of hypothesized but as-yet-unobserved phenom-ena, such as primordial black hole evaporation and prompt emission associated with coalescingmassive objects have been suggested. The 2007 announcement by Lorimer et al. of the detectionof a bright (30 Jy) radio pulse that was inferred to be of extragalactic origin and the subsequentconsternation have demonstrated both the potential utility of bright radio pulses as probes of theinterstellar medium and intergalactic medium, as well as the need for wide-field surveys character-izing the fast-transient parameter space. Here we present results from the 450 hour, 150 deg Fly’sEye survey for bright dispersed radio pulses at the Allen Telescope Array (ATA). The Fly’s Eyespectrometer produces 128 channel power spectra over a 209 MHz bandwidth, centered at 1430MHz, on 44 independent signals paths originating with 30 independent ATA antennas. Data werededispersed between 0 and 2000 pc cm − and searched for pulses with dispersion measures greaterthan 50 pc cm − between 625 µ s and 5 s in duration. No pulses were detected in the survey, imply-ing a limiting rate of less than 2 sky − hour − for 10 millisecond duration pulses having apparentenergy densities greater than 440 kJy µ s, or mean flux densities greater than 44 Jy. Here we presentdetails of the instrument, experiment and observations, including a discussion of our results in lightof other single pulse searches.
1. Introduction
The last decade has seen an explosion of interest in time domain radio astronomy. Drivenby new wide field survey instruments, multi-beam receivers and computing advances, explorationof this regime presents opportunities to shed new light on known phenomena and perhaps revealpreviously unseen processes as well (Cordes 2007). Investigations in time domain radio astronomycan be conveniently divided into two categories: those that deal with slow transients, events lasting University of California, Berkeley Oxford University Stanford University Cornell University Netherlands Institute for Radio Astronomy (ASTRON) a r X i v : . [ a s t r o - ph . H E ] S e p − . Even liberal models for the galactic and SMC contribution to the total impliedelectron column density could account for only a fraction of the DM measured. Assuming the restof the dispersion was due to a Milky Way-like host interstellar medium (ISM) contribution and 3 –traversal of the much more rarified intergalactic medium (IGM), the lower limit on the distanceto the source was calculated to be ∼
600 Mpc. Suffice it to say, the implied energy release of ∼ ergs presented a challenge for astrophysical theory, and motivated wide ranging speculationof possible origins. Several subsequent searches (Keane et al. 2010; Deneva et al. 2009) did notdetect any similar events, implying that such events must be exceedingly rare. Burke-Spolaoret al. (2010) presented the detection of several additional impulsive events in Parkes survey datawith similar dispersive chirps to that seen in the Lorimer Burst but exhibiting clear indications ofterrestrial origins. While the Burke-Spolaor et al. (2010) events showed an approximately quadraticfrequency evolution, as would be expected for an astrophysical event, the detection of the eventsin multiple receiver beams simultaneously clearly points to a terrestrial source and the irregularityof received flux across the observing band and large pulse width differentiate them markedly fromthe Lorimer Burst. Recently, another possibly extragalactic burst was discovered in additionalre-analysis of Parkes survey data (Keane et al. 2011), lending some support for the existence of abonafide population of very bright extragalactic fast transient sources. Regardless of the sourceof such bursts, a population of extragalactic objects or events producing extraordinarily energeticradio pulses would provide an invaluable probe of the ionized IGM.Here we present a search using the 42-dish Allen Telescope Array for bright dispersed radiopulses, with specific attention paid to those of possible extragalactic origin. Section 2 describesthe digital spectrometer developed for this experiment, installation, verification and calibrationprocedures, Sections 3 and 4 detail observations and analyses and Section 5 presents our resultsand interpretation.
2. Instrument and Installation
The single pulse search described here used the Allen Telescope Array (ATA) (Welch et al.2009) in an unconventional non-interferometric mode. Rather than pointing all 42 dishes in thesame direction, each dish was pointed at a unique position, similar to a “fly’s eye.” Such a modeyields a dramatically increased field of view at the expense of sensitivity, well matched to detectingbright, rare events. The primary half-power beam width (HPBW) of the ATA is approximately 2.5deg at 1.4 GHz, yielding a potential field of view of more than 200 deg .In this experiment, each antenna signal path was processed independently using a purpose-built digital spectrometer. This device, dubbed the Fly’s Eye Spectrometer, was constructed usingthe modular instrumentation infrastructure developed by the Center for Astronomy Signal Pro-cessing and Electronics Research (Werthimer et al. 2011). The full system consists of eleven fieldprogrammable gate-array (FPGA)-based ‘iBOB’ computing boards, each equipped with two 1024Msample/sec ‘iADC’ analog-to-digital converter cards. Each ADC board digitizes two independentsingle-polarization signal paths at 838.8608 Msamples/sec. The Nyquist sampled band is digitallydown converted and decimated to a bandwidth of 209.7152 MHz within the FPGA and channelizedusing a 2 channel complex biplex-pipelined polyphase filterbank. Power spectra are detected andaccumulated for 625 µ s, packetized into Ethernet UDP packets on each of the eleven FPGA boardsand transmitted to a single Linux PC via an Ethernet switch. The payload of each UDP packet 4 –contains a 21 byte header, which specifies a board ID, accumulation number and any error condi-tions, followed by 512 bytes of spectral data (unsigned byte power measurements ×
128 frequencychannels × . This level of error translates to a center RF frequency and bandwidth ambiguity of about84 kHz. Because the Fly’s Eye spectrometer derives integration time from counting the samplingclock, the unlocked clock synthesizer also imposes an ambiguity in integration time of about 60nanoseconds. In total, these effects correspond to an additional ∼ S / N = S mean S sys (cid:112) n p t int ∆ f N h (1)where S / N is the detected signal to noise ratio, S mean is the mean flux density of the pulsar, S sys isthe system equivalent flux density (SEFD) of the observing system, n p the number of polarizationssummed, t int the integration time, ∆ f the bandwidth observed and N h is the number of harmonicsused in a harmonic sum, which depends on the pulse period P and pulse width W as N h ≈ P/W .For the case of multiple antennas sampled synchronously and detected independently, n p is equal 5 – S p ec t r o m e t e r B o a r d ATA Fly’s EyePSR B0329+54 - 12/2007
Signal Path
A B C D
Fig. 1.—: Folded pulse profiles of PSR B0329+54 as observed in individual Fly’s Eye Spectrometerinputs for a 1290s integration at 1430 MHz. Two full turns plotted for clarityto the total number of antenna-polarization signals incoherently summed, n ant − pol . Applying thisequation to the summed profile shown in Figure 2 with n ant − pol = 44 and the expected meanSEFD for individual ATA antennas ( ∼
10 kJy), we infer a flux density for B0329+54 of S mean ≈ S mean = 190 mJy, especially considering that the pulsar B0329+54 has a variable meanobserved flux density of a factor of ∼ P ( E i > E ) = KE α , where P gives the probability of a pulse having a pulse area E i greater than E . Here pulse area is defined as E i = S i W i , where S i is the mean intrinsic flux density of the pulse over an intrinsic time W i . Whileother authors have referred to the quantity E i as “energy”, in later portions of this work we will 6 – Pulse Phase 110º - 250º F l ux ( A r b it r a r y ) Fly’s Eye Gould and Lyne, 1998
Fig. 2.—: PSR B0329+54 folded pulse profile as detected by the incoherent sum of all 44 inputs tothe Fly’s Eye Spectrometer (shown in 1) for a 1290s integration at 1430 MHz (left) and a referenceprofile from Gould & Lyne (1998) taken at 1408 MHz.use the slightly more accurate “energy density.” At 1300 MHz, Bhat et al. (2008) gives α ∼ -1.9for energy densities greater than 10 kJy µ s, with K = 4 . × − and E in kJy µ s. Rearrangingequation 3 in Deneva et al. (2009) gives an expression for the minimum detectable energy density, E i = mS sys √ W (cid:112) n p ∆ f (2) W is the observed pulse width, usually taken to be the quadrature sum of the various sources ofbroadening, both astrophysical and instrumental and m the signal-to-noise threshold. For a pulsewith an observed width limited by our digital hardware and assuming a single polarization SEFDof ∼
10 kJy, a characteristic m = 5 σ minimum detectable energy density for the incoherent sum of19 inputs is 20 kJy µ s or an mean flux density of ∼
32 Jy for a 625 µ s pulse. Based on the expectedCrab GP distribution, we should observe a pulse with E i >
20 kJy µ s about 18 times per houron average. Figure 3 shows the detection of 10 bright GPs at the expected DM of 56.8 pc cm − for the Crab pulsar in a 50 minute unweighted incoherently summed observation using the 19 bestperforming FE inputs, as determined from Figure 1.The set of 44 antenna inputs ultimately used for the FE observing campaign originated with 30independent antennas, 14 of which included both X and Y polarizations. Figure 4 shows antennaperformance, described by the system equivalent flux-density (SEFD), for each of the 44 inputschosen for the Fly’s Eye observing campaign. The values given here were determined by interfer-ometric observation of standard calibrators interleaved with Fly’s Eye observation. Details of theFly’s Eye instrument parameters are given in Table 1.
3. Observations
During the period February 2008 to May 2008, we conducted approximately 480 hours ofdrift-scan observations with the Fly’s Eye Spectrometer ( Table 2). Data were collected in 60minute intervals, each consisting of 58 minutes of drift observation followed by 2 minute diagnosticobservations used for monitoring the health of the telescope and instrumentation. Fly’s Eye obser-vations produced data at a rate of roughly 36 GB / hour, resulting in approximately 18 TB totalcollected data for the entire observation period. The data are archived in Berkeley and available 7 –
Source: Crab Pulsar
MJD: 54456.418 Max SNR: 23.8
Instrument: Fly’s Eye F ctr (cid:29)(cid:3)(cid:20)(cid:23)(cid:22)(cid:20)(cid:17)(cid:25)(cid:3)(cid:48)(cid:43)(cid:93)(cid:3) (cid:3) (cid:520)(cid:3)(cid:44)(cid:81)(cid:83)(cid:88)(cid:87)(cid:86)(cid:29)(cid:3)(cid:20)(cid:28) Fig. 3.—: A standard single pulse detection plot for a ∼
50 minute observation of the Crab pulsarafter summing 19 Fly’s Eye inputs. Several giant pulses are apparent at a DM of 56.8 pc cm − .Other features are radio frequency interference. From top, left to right, the panels show a histogramof detection signal-to-noise ratio (SNR) for pulses > σ , a histogram of detection dispersion measure,detection dispersion measure against detection SNR and detection time vs. dispersion measurewith detection signal to noise indicated by plot point radius. This plot produced using PRESTO(Ransom 2001).for analysis by request to the authors. At the time of these observations, the ATA was undergoingcommissioning, and a variety of system performance issues were being actively addressed. Thevariation in SEFD from antenna to antenna and less-than-complete utilization of the 42 installedantennas are reflective of these issues. To aid in dynamically determining signal path operability,a fixed pointing strip along a constant declination angle of +54 ◦ , in which antennae were spaced1 half-power beam width apart, was chosen for drift scan observations. As the bright pulsar PSRB0329+54 drifted through the beam pattern at the sidereal rate, its detection or non-detectionwas used to determine whether or not a particular signal path was operable. Declination +54 ◦ iswell away from any significant source of galactic electron density confusion, the median maximumgalactic DM contribution along this path is ∼
68 pc cm − (from the NE2001 model, Cordes &Lazio (2002)). Figure 5 illustrates the overall observing efficiency after applying the B0329+54detectability metric. Out of a total of 921.1 input-days of observing, 579.9 input-days showed theexpected detections of B0329+54, for a total observing efficiency of ∼ ◦ . Although we believe most signal paths were operable, we have conservativelyexcluded these data. 8 – SEFD (kJy) A n t e nn a / S p ec t r o m e t e r I npu t Fig. 4.—: Mean antenna-polarization performance for each input to the Fly’s Eye Spectrometer,as determined by interferometric observation of standard calibrators interleaved with Fly’s Eyeobservations between 02/2008 and 05/2008. Errors are ± σ .
4. Analysis4.1. Data Preparation
Power spectra time series for each of the 44 inputs to the Fly’s Eye Spectrometer were extractedas individual “filterbank” format files (Lorimer et al. 2000), broken into analysis chunks of length2 samples (representing 12 minutes). This length was chosen to allow an entire analysis chunk andset of dedispersed time series to be kept in computer memory during analysis. Prior to dedispersion,power spectra were normalized or “equalized” across both frequency channel and time. Equalizingacross frequency channels has the primary effect of correcting for the rippled bandpass imposedby both analog filter response and digital down conversion. Equalization of accumulation valuesmitigates broadband gain changes and broadband impulsive interference. This process was carriedout as follows.We denote the power in channel i at (discrete) time t as P i ( t ) ∈ [0 , Center Frequency ν o Number of Channels N chan Channel Width ∆ ν i Bandwidth ∆ ν
210 MHz
Beam Width
Θ 2.5 deg a (per antenna, HPBW) Solid Angle
Ω 147.3 deg b (Instantaneous) System Temperature c T sys
50 K
Gain c G . × − K/Jy
Dish Diameter D ∼ a MacMahon & Wright (2009) b Assuming all signal paths are operable butaccounting for some dual-polarization observations, seeSection 2 c Nominal value
Table 2:: Fly’s Eye Observations 02/2008 − Epoch MJD Efficiency1 mean power per channel P i ≡ T T − (cid:88) t =0 P i ( t ) (3)over some time period T . T is typically set to the length of an analysis chunk, 2 samples. Aparticular value P i ( t ) is then divided by the mean P i . i.e. the equalized value P (cid:48) i ( t ) ≡ P i ( t ) /P i .The mean power in each channel is then unity, since1 T T − (cid:88) t =0 P (cid:48) i ( t ) = 1 (4) 10 – Epoch A n t e nn a / S p ec t r o m e t e r I npu t Fig. 5.—: Diagram showing operability of each antenna/input as a function of epoch, based ondetectability of B0329+54. Filled circles indicate that a given signal path is operable. For aninput/epoch pair to be considered operable, B0329+54 must have been detected at every opportu-nity within the epoch. On the Y axis are each spectrometer input, labeled by their ATA antennaidentifier followed by spectrometer input number. The suffix on the antenna identifier indicateswhich of two dual linear polarization feeds was used.Mean power equalization was performed on the frequency spectrum equalized values P (cid:48) i ( t ). Thepower mean over all frequency channels for a single integration (time sample t ) is defined as P (cid:48) ( t ) ≡ N N − (cid:88) i =0 P (cid:48) i ( t ) (5) N is the number of channels. With the mean powers P (cid:48) ( t ), we can define the equalization of thepowers P (cid:48) i ( t ). The mean power equalized values P (cid:48)(cid:48) i ( t ) ≡ P (cid:48) i ( t ) /P (cid:48) ( t ). This procedure ensures thatthe sum of the power samples for any time T is normalized such that (cid:80) N − i =0 P (cid:48)(cid:48) i ( t ) = N , effectivelyflattening the DM = 0 time series.Prior to the equalization process, individual frequency channels with especially large amounts 11 –of interference were identified and logged. Our algorithm used the variance of each frequencychannel over an analysis chunk as a measure of the amount of interference in that channel. Usingthe previously defined quantities, we computed the variance of the values P i ( t ) for each channel i over T , and then fit a polynomial to the resulting curve Var( P i ) (Figure 6). Frequency channelsfor which the computed variance differed from the polynomial fit by Var( P i ) > σ were excludedfrom subsequent dedispersion. We explicitly excluded 8 frequency channels at the top of the bandand 13 frequency channels at the bottom of the band due to analog filter roll off and the presenceof bright air route surveillance radar below 1350 MHz. Frequency Channel
VariancePolynomial Fit (cid:14)(cid:18)(cid:16)(cid:3)(cid:22)(cid:433)
Excluded Frequency ChannelsA Priori Excluded Regions (cid:433)
32 64 96 128 Fig. 6.—: Variance vs. frequency channel for a 2 spectra Fly’s Eye observation (red), an iterativepolynomial fit to the variance curve (green) and ± σ bounds on the fit (black). Band edges areexcluded a priori from the polynomial fit, as instrumental response is poor in these regions due tofilter roll-off. Band edges and individual frequency channels with high variance are excluded fromthe de-dispersion process. A single pulse search in dedispersed time series for events having a signal-to-noise ratio (SNR)greater than five standard deviations above the mean, σ > .
0, was carried out over 744 trialDMs between 0-2002 pc cm − using the SigProc tools (Lorimer et al. 2000). An approximatematched filtering algorithm was employed to increase sensitivity to broadened pulses in which eachdedispersed time series was iteratively smoothed by adding 2 n adjacent time samples over the rangen = 0 to 10, following Cordes & McLaughlin (2003). This results in an effective box car smoothingof maximal window size 2 samples or 0.64 to 5.12 s, depending on the level of time collapse (Table3). The initial DM step size was set such that the time delay associated with the DM step wasless than the sampling interval. Spectra were iteratively collapsed in time by a factor of two atDMs 329.0, 658.0, 1314.0 pc cm − to speed analysis, as detailed Table 3. At these thresholds apulse would subtend a minimum of 2, 4 or 8 spectra at the top of the band, respectively, and thushalving the effective time resolution imposes no loss in sensitivity. Similarly, the DM step size canbe reduced by the same factor and remain less than the new effective time resolution. These searchparameter choices were all designed to ensure that the dominant source of temporal smearing was 12 –due to the unavoidable (at the time of analysis) integration time smearing and in-band dispersivesmearing. Like many problems in radio astronomy, the analysis of multibeam pulse search data isreadily parallelizable. Here we distributed 58 minute observations to individual compute nodes, andparallelized inputs over individual CPU cores. In total, our analysis consumed ∼ Dispersion Measure Range ∆ DM Time Collapse Factor − After dedispersion, any strong signal present in a dynamic spectra – be it from interferenceor a real event – will be detected at multiple DMs, strengths and times. The distribution ofthese detections depends on the observed properties of the signal, the range of DMs searched andany pre- or post-dispersion processing applied. In the case of quadratically chirped radio pulses,this distribution follows a characteristic functional form (Cordes & McLaughlin 2003). Likewise,certain kinds of interference will exhibit predictable detection distributions. Wideband, temporallynarrow RFI will be detected over many DMs with the highest strength detections at low DMs.Narrowband, long duration RFI will also be detected at many DMs but will peak in strength whenthe dispersion path for a trial dispersion measure optimally overlaps the narrow-band interference.Finally, wideband and long duration interference or rapid gain changes will cause an excess ofdetections in all DMs for the duration of the event.As a first cut on the vast number of high SNR candidates detected, we flagged any 1 secondinput-time region where the highest SNR candidate in that region was <
50 pc cm − or > − , or the total number of pulses over 5 σ in the same DM regimes exceeded a factor of 4 timesthe mean number of pulses in each regime. The DM <
50 pc cm − would reject any relativelynearby galactic events, but distinguishing true astrophysical bursts from interference at these lowDMs is very difficult because of the correspondingly small quadratic chirp. Further, our focuswas primarily on potentially extragalactic events for which the DM contribution from the MilkyWay and host ISM should well exceed this threshold. Figure 7 shows the results of applying thismetric to pulse detections from two observations of the Crab Pulsar. The algorithm was effectiveat rejecting strong interference and avoided rejecting true astrophysical events. Strong interferencedominates Figure 7a, but is greatly diminished in 7a. Figures 7b and 7d shows the detection of abright pulse left untouched after applying our algorithm.Following RFI rejection, all events with a SNR σ > . sample analysis chunksfor which the mean SNR of all detections was σ < . Crab Pulsar Observation: Single Pulse Detections (Raw)
Crab PulseDM ~ 58
Time (seconds) D i s p e r s i on M ea s u r e ( p c c m - ) (a) Crab Pulsar Observation: Single Pulse Detections (Raw)
Crab PulseDM ~ 58
Time (seconds) D i s p e r s i on M ea s u r e ( p c c m - ) (b) Crab Pulsar Observation: Single Pulse Detections (RFI Removed)
Crab PulseDM ~ 58
Time (seconds) D i s p e r s i on M ea s u r e ( p c c m - ) (c) Crab Pulsar Observation: Single Pulse Detections (RFI Removed)
Crab PulseDM ~ 58
Time (seconds) D i s p e r s i on M ea s u r e ( p c c m - ) (d) Fig. 7.—: Detection of giant pulses from the Crab pulsar for two individual FE inputs before (7a,7b) and after (7c, 7d) application of a post processing RFI filter. Each plot shows events vs. timeand trial dispersion measure, here the radii of plot points are proportional to (signal to noise)
14 –extracted and closely examined t vs. ν spectrograms of 1, 2, 4 and 10 seconds around the event.Figure 8 shows the SNR distributions for all pulses detected in operable signal paths, all pulses inoperable paths after applying the first cut RFI rejection algorithm described above, and (inset) allpulses detected in observations having a mean pulse detection SNR σ < .
5. Upon close inspection,none of the pulse candidates identified appeared to be of astronomical origin. An example of thepathological interference that escaped our interference rejection algorithms is shown in Figure 9.Strong pulses were detected in regions where dedispersion curves aligned with the triangle-wavemodulation of the interferer, at a DM of ∼
80 pc cm − . This particular interferer was detected atmultiple epochs, and appears to originate with an orbiting satellite.
50 100 150 200 250 30010 D e t ec t i o n C o un t Signal to Noise Ratio A ll O p er a b l e D a t a I n i t i a l R F I R e j ec t i o n Inset
Final RFI Rejection
Fig. 8.—: SNR histograms for all pulses detected in operable signal paths, all pulses in operablepaths after applying the first cut RFI rejection, and (inset) all pulses detected in 2 sampleobservation chunks having a mean pulse detection SNR σ < .
5. Discussion
Our results indicate that the millisecond radio sky is relatively quiescent at the energies probedby our experiment. For a threshold of 8 σ , our minimum detectable energy density is E min = 111kJy µ s or a mean flux density of ∼
178 Jy for a 625 µ s pulse. Based on our non-detection, pulsesof these energy densities originating with an isotropic progenitor must occur at a rate less than 2sky − day − . We can determine the limiting rate of occurrence of bursts as a function of antennasensitivity by computing the rate η ( E > E min ) = 1 (cid:80) Ω · T obs ( SEF D > SEF D min ) (6)where E min corresponds to the detectable limit for pulses of a given duration in an antenna having SEF D min . Figure 11 shows the event rate limit calculated from these results alongside other 15 – F r e qu e n c y ( M H z ) Time (seconds) Time (milliseconds) P o w e r ( A r b it r a r y , L i n ea r) Fig. 9.—: A spectrogram plot of intensity in the time-frequency plane of a satellite interferer thatpassed our RFI rejection algorithms. Edges of the observed band, where system response is poor,have been masked.recent single pulse searches in Table 4. Our rate limit does not yet sample potential coherent radioemission processes from neutron star binary inspirals or gamma ray bursts, but we can place anorder of magnitude limit on the luminosity of coherent radio emission from core collapse supernovae(CC SNe). Assuming isotropic emission, our survey could have detected ∼
20 10 ms events withan intrinsic rate of 1000 sky − day − at an apparent energy density limit of E limit = 10 kJy µ s.Assuming that coherent radio emission from CC SNe is beamed over a solid angle Ω, we can limitthe emission cone to Ω < π/
20. Translating this into an upper limit on luminosity using the radiusof a spherical volume of 1 Gpc , we have L < E limit ∆ ν πR
20 1∆ t (7)For emission over a bandwidth ∆ ν = 1 GHz and ∆ t = 10 ms we have L < ∼ × erg sec − .As discussed in Lorimer et al. (2007) and Keane et al. (2011), if the two detected pulses inferredto be at cosmological distances are indeed real, they must represent an entirely new source class.The extreme SNR of the two sparse detections is curious, and as previously discussed by severalauthors, contradicts the assumption of an isotropically distributed population. In the case of thisexperiment, we would have expected to detect ∼
15 events similar to the Lorimer et al. (2007)burst, but (cid:28) The incompatibility ofthe implied rates for the two isolated detections is difficult to explain. The galactic latitude of thetwo events differ significantly, b = − . ◦ for the Lorimer et al. (2007) event and b = − ◦ for theKeane et al. (2011) burst. While one might guess that being much closer to the galactic planewould make it easier to explain a large dispersion measure, of the 227 known pulsars between 3 ◦ and 5 ◦ off the galactic plane, none has a DM >
460 pc cm − , standing in stark contrast with the Assuming an Euclidean isotropic distribution and correcting for the Fly’s Eye survey’s lower sensitivity. See, forexample, Deneva et al. (2009) for an exposition of this calculation.
16 –Keane et al. (2011) event at a DM of 745 pc cm − However, the increased intragalactic path lengthof this detection does present more opportunity for an unseen highly ionized nebula to make anaberrant contribution to the total integrated free electron column density.The 15 Burke-Spolaor et al. (2010) detections also remain puzzling. Assuming the phenomenathat generated these bursts is not unique to the Parkes site, we can estimate the number of similarevents that the Fly’s Eye survey should have detected. Assuming that these bright events wereobserved in far out side lobes, a reasonable comparison between the two surveys reduces to simplythe ratio of their observing time, thus the Fly’s Eye survey should have detected15 events × . . ∼
10 events (8)of these events as well. Note that here we assume antenna efficiency and system temperaturedifferences are negligible. We will refrain from speculating on the cause of these events, but notethat our observations well sampled the diurnal cycle as well as varying levels of precipitation. Weare unaware of any lightning activity in the near vicinity of the ATA during our observations.It remains perplexing as to why all of these unique transient bursts have been detected only atthe Parkes telescope. However, with the aggressiveness with which interference must be excised insuch experiments, and the varying means by which it is accomplished, we speculate that it is notwholly out of the question that some processing pipelines could be better tuned to detecting singlepulses at extragalactic DMs or other unexpected characteristics. The fact that all of these pulseswere themselves discovered in reanalyses of previously mined surveys, with the second extragalacticevent discovered in a 3rd reanalysis, indicates that other extant surveys may harbor additionalas-yet undetected events. Our own pipeline has been honed by the exercise of this experiment,and a future search would undoubtedly be superior. For instance, summing polarizations wouldyield an additional √
6. Summary
We have developed a novel multiple-input digital spectrometer which we have used to conducta wide field search for bright dispersed radio pulses using the Allen Telescope Array. This widefield search yielded no detections, allowing us to place a limiting rate of less than 2 sky − hour − for 10 millisecond duration pulses having mean apparent flux densities greater than 44 Jy. Theflux densities probed by this experiment are well above individual pulses from known pulsars andRRATs, just grazing the very brightest of the giant pulse producing pulsars, none of which arepresent in the field surveyed. We have placed new limits on very bright coherent emission fromevents similar to the singular event described in Lorimer et al. (2007). Our results indicate thatthe Lorimer et al. (2007) event must belong to a very rare source class, if it is indeed astrophysical.We did not detect any quadratically dispersed terrestrial interference similar to that seen at theParkes observatory, e.g. Burke-Spolaor et al. (2010), consistent with other non-Parkes surveys, e.g.Deneva et al. (2009).The work presented here has shown that sources of bright fast transient radio emission mustbe relatively rare. We have also shown both the utility, and associated of challenges, of using aninterferometric array in a multi-pointing “fly’s eye” mode. The next generation of radio inter-ferometers currently being built in preparation for the Square Kilometer Array (Dewdney et al.2009) will offer new opportunities to explore the fast transient regime with greater sensitivity overlarge solid angles. The use of these new instruments for fly’s eye mode surveys will require mak-ing individual antenna or station data accessible to sufficient digital hardware, and we encouragethis consideration to be taken into account during the design phase. Commensal surveys for fasttransient emission with interferometers, using the incoherent sum of antennas pointed in the samedirection, could offer an excellent trade-off between sensitivity and solid angle while incurring littleadditional hardware cost and no additional observing time, see e.g. Macquart (2011). Again, suchcapabilities will require consideration early in the design process in order to be realized efficiently.The exploration of the fast radio transient parameter space is just beginning, and the contrastingresults of this and other experiments clearly indicate we have much yet to learn.
7. Acknowledgements
The ATA is jointly operated by the University of California, Berkeley Radio Astronomy Laband the SETI Institute in Mountain View, California. The authors would like to acknowledgethe generous support of the Paul G. Allen Family Foundation, who have provided major supportfor design, construction, and operations of the ATA. Contributions from Nathan Myhrvold, XilinxCorporation, Sun Microsystems, and other private donors have been instrumental in supporting the 18 –ATA. The ATA has been supported by contributions from the US Naval Observatory in additionto National Science Foundation grants AST-050690 and AST-0838268.We thank Scott Ransom, Matthew Bailes, Eric Korpela and Josh von Korff for productivediscussions and Garrett Keating for providing the interferometric standard calibrator observationsfrom which our sensitivity estimates are derived. We also thank the Collaboration for AstronomySignal Processing and Electronics Research and its members for contributing to the instrumentationdevelopment framework used in construction of the Fly’s Eye spectrometer. We thank Henry Chen,Jeff Cobb, Matt Dexter, Terry Filiba, Rick Forster, Colby Gutierrez-Kraybill, Matt Lebofsky,David MacMahon and Jason Manley for invaluable engineering and data management support andexpertise.We also thank an anonymous reviewer for a careful reading of an earlier draft of this manuscriptand thoughtful comments which have improved this work.
SEFD (kJy) S o li d A ng l e (cid:117) T i m e ( d e g (cid:117) hou r s ) OperableInterference Rejected
Fig. 10.—: A plot of system sensitivity, described by the system equivalent flux density, vs. totaloperable observing, described by the observing solid angle · time product. Plot includes bothoperable observing, taken from the set of signal paths where PSR B0329+54 was consistentlydetected, and the final low-RFI analysis set. In cases where two polarizations were observed for agiven antenna, the lower SEFD was used. 19 – (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) L o r i m er e t a l Deneva et al 2010 (primary beam)
Keane et al 2011Burke-Spolaor et al 2010 Siemion et al 2011 (off axis) “Lorimer Burst” (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s) Single Detection (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s) Upper LimitBright Extragalactic Radio Pulse
Siemion et al 2011 (This Work)
Gamma Ray Bursts(per Gpc )Binary Neutron Star Inspirals(per Gpc )Core Collapse Supernovae(per Gpc ) Parkes “Terrestrial Events” (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s)10 (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s)10 (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s)10 (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s)10 (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s)10 (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s)10 (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s)10 (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s)10 (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s)10 (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s)10 (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s)10 (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s)10 (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s)10 (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s)10 (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s)10 (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s)10 (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s)10 (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s)10 (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s)10 (cid:239)(cid:21) (cid:239)(cid:20) (cid:21) (cid:239)(cid:20) (cid:21) E v e n t R a t e ( s k y (cid:239) (cid:20) d ay (cid:239) (cid:20) ) Pulse Energy Density (kJy (cid:82) s) Deneva et al 2010 (off axis)
Fig. 11.—: Pulse energy density vs. rate limit for the surveys in Table 4. Rate limit curves assumea 10 millisecond pulse duration. Shaded bars on the Lorimer Burst and the extragalactic eventdescribed in Keane et al. (2011) represent 2 sigma confidence (Gehrels 1986) on the Euclideanisotropic distribution. The point for terrestrial events identified in Burke-Spolaor et al. (2010)assumes off-axis detection. Rates of core collapse supernovae, gamma ray bursts and binary neutronstar inspirals in a 1 Gpc volume are taken from Madau et al. (1998), Guetta & della Valle (2007)and Kalogera et al. (2004), respectively, via Lorimer et al. (2007). Here we assume flat sensitivityacross the HPBW of each receiver beam and do not take into account reduced sensitivity in widefield-of-view sidelobes, except in the case of this work and Deneva et al. (2009). The curved linefor the Siemion et al survey reflects variation in antenna system temperature. 20 –Table 4:: Recent L-band Single Pulse Surveys Survey T obs
Total Solid Angle E min a (hours) (deg ) kJy µ sEdwards et al. (2001)Re-analysis by Burke-Spolaor & Bailes (2010) 346.1 0.556 b c d ea For a 10 millisecond pulse. Values given here are conservative, see references for details. b Parkes Multibeam - 14 (cid:48)
HPBW/beam ×
13 beams c Arecibo ALFA - 3.5 (cid:48)
HPBW/beam × d ◦ HPBW/beam ×
30 beams e Using a nominal SEFD of 8 kJy
21 –
REFERENCES
Backer, D. C. 1970, Nature, 228, 42Becker, R. H., Helfand, D. J., White, R. L., & Proctor, D. D. 2010, The Astronomical Journal,140, 157Benford, G., Benford, J., & Benford, D. 2008, arXiv:0810.3966v3 [physics.pop-ph]Bhat, N. D. R., Tingay, S. J., & Knight, H. S. 2008, The Astrophysical Journal, 676, 1200Bower, G. C., & Saul, D. 2011, The Astrophysical Journal Letters, 728, L14Bower, G. C., Saul, D., Bloom, J. S., Bolatto, A., Filippenko, A. V., Foley, R. J., & Perley, D.2007, The Astrophysical Journal, 666, 346Burke-Spolaor, S., & Bailes, M. 2010, Mon. Not. R. Astron. Soc., 402, 855Burke-Spolaor, S., Bailes, M., Ekers, R., Macquart, J.-P., & III, F. C. 2010, arXiv:1009.5392v1[astro-ph]Cordes, J. 2007, SKA Memo, 97Cordes, J., Bhat, N., Hankins, T. H., McLaughlin, M., & Kern, J. 2004, The Astrophysical Journal,612, 375Cordes, J. M., & Chernoff, D. F. 1998, The Astrophysical Journal, 505, 315Cordes, J. M., & Lazio, T. J. W. 2002, arXiv:0207156v3 [astro-ph]Cordes, J. M., & McLaughlin, M. A. 2003, The Astrophysical Journal, 596, 1142Deneva, J. S., et al. 2009, The Astrophysical Journal, 703, 2259Dewdney, P. E., Hall, P. J., Schilizzi, R. T., & Lazio, T. J. L. W. 2009, Proceedings of the IEEE,97, 1482Edwards, R. T., Bailes, P. M., Straten, W. V., & Britton, M. C. 2001, Mon. Not. R. Astron. Soc.,326, 358Fridman, P. 2010, Mon. Not. R. Astron. Soc., 409, 808Gehrels, N. 1986, Astrophysical Journal, 303, 336Gould, D. M., & Lyne, A. G. 1998, Mon. Not. R. Astron. Soc., 301, 235Guetta, D., & della Valle, M. 2007, The Astrophysical Journal, 657, L73Hallinan, G., Antonova, A., Doyle, J. G., Bourke, S., Lane, C., & Golden, A. 2008, The Astrophys-ical Journal, 684, 644Hansen, B. M. S., & Lyutikov, M. 2001, Mon. Not. R. Astron. Soc., 322, 695 22 –Hogden, J., Wiel, S. V., Bower, G. C., Michalak, S., Siemion, A., & Werthimer, D. 2011, TheAstrophysical Journal, SubmittedJacoby, B. A., Bailes, M., Ord, S. M., Edwards, R. T., & Kulkarni, S. R. 2009, The AstrophysicalJournal, 699, 2009Kalogera, V., et al. 2004, The Astrophysical Journal, 601, 179Keane, E. F., Kramer, M., Lyne, A. G., Stappers, B. W., & McLaughlin, M. A. 2011, Mon. Not.R. Astron. Soc., In PressKeane, E. F., Ludovici, D. A., Eatough, R. P., Kramer, M., Lyne, A. G., McLaughlin, M. A., &Stappers, B. W. 2010, Mon. Not. R. Astron. Soc., 401, 1057Liu, Z. Y., Wang, N., Urama, J. O., & Manchester, R. N. 2006, Chinese Journal of Astronomy andAstrophysics, 6, 64Lorimer, D., & Kramer, M. 2005, Cambridge University PressLorimer, D. R., Bailes, M., McLaughlin, M. A., Narkevic, D. J., & Crawford, F. 2007, Science, 318,777Lorimer, D. R., Kramer, M., M¨uller, P., Wex, N., Jessner, A., Lange, C., & Wielebinski, R. 2000,Astronomy and Astrophysics, 358, 169MacMahon, D., & Wright, M. 2009, ATA Memo Series, 80Macquart, J.-P. 2011, The Astrophysical Journal, 734, 20Madau, P., della Valle, M., & Panagia, N. 1998, Mon. Not. R. Astron. Soc., 297, L17Manchester, R. N., Fan, G., Lyne, A. G., Kaspi, V. M., & Crawford, F. 2006, The AstrophysicalJournal, 649, 235Manchester, R. N., Hobbs, G. B., Teoh, A., & Hobbs, M. 2005, The Astronomical Journal, 129,1993Manchester, R. N., et al. 2001, Mon. Not. R. Astron. Soc., 328, 17McLaughlin, M. A., et al. 2006, Nature, 439, 817McMahon, P. L. 2008, arXiv:1109.0416 [astro-ph.IM], M.Sc. Thesis, University of Cape TownOfek, E. O., Breslauer, B., Gal-Yam, A., Frail, D., Kasliwal, M. M., Kulkarni, S. R., & Waxman,E. 2010, The Astrophysical Journal, 711, 517Ransom, S. M. 2001, Harvard University PhD ThesisRees, M. J. 1977, Nature, 266, 333Siemion, A., et al. 2010, Acta Astronautica, 67, 1342 23 –Vachaspati, T. 2008, Physical Review Letters, 101, 141301Wang, N., Manchester, R. N., Johnston, S., Rickett, B., Zhang, J., Yusup, A., & Chen, M. 2005,Mon. Not. R. Astron. Soc., 358, 270Welch, J., et al. 2009, Proceedings of the IEEE, 97, 1438Werthimer, D., Backer, D., Wawrzynek, J., & Wright, M. 2011, http://casper.berkeley.edu
Zarka, P. 2007, Planetary and Space Science, 55, 598Zarka, P., Farrell, W., Fischer, G., & Konovalenko, A. 2008, Planetary Atmospheric Electricity,137, 257