A Survey for Transient Astronomical Radio Emission at 611 MHz

Many sources have been observed to produce transient astronomical radio emission. Perhaps the most familiar is the Sun, from which a large variety of transient radio signals emanate. From microwave spike bursts lasting ∼10 ms to type III storms lasting weeks, the characteristics of solar radio bursts span wide ranges of brightness temperature, duration, frequency, and polarization. Associated with many different aspects of solar activity, the mechanisms producing the bursts run the gamut, from thermal bremsstrahlung to plasma radiation. Dulk (1985) and Hjellming (1988) provide reviews. Other stars have been seen to produce radio bursts as well. Events observed on flare stars are similar in character to those seen on the Sun but imply radio luminosities 10⁴ times that of solar flares (Dulk 1985). RS CVn systems, close binaries with orbital periods of ∼1–30 days, have been observed to produce transient radio emission on timescales of minutes to days. The signals are thought to be due to two separate phenomena: the acceleration of electrons in the magnetic fields between the stars and coherent emission from the individual stars (Hjellming 1988). X‐ray binaries as well are known to produce transient radio signals following X‐ray events, although the mechanism in this case is thought to be quite different from that of the RS CVn binaries (Hjellming & Han 1995). Jupiter was discovered many years ago to produce transient radio emission at decameter wavelengths (Burke & Franklin 1955), thought to be due to synchrotron radiation from high‐energy electrons trapped in the magnetic field of the planet. Details are reviewed in Carr, Desch, & Alexander (1983). Brown dwarfs have been seen to flare in the radio: a recent VLA observation of a such a flare measured a flux density much higher than that expected from an empirical relation between the luminosities of brown dwarf radio flares and X‐ray flares (Berger et al. 2001). Some radio pulsars are observed occasionally to produce "giant pulses." For example, about 0.3% of the pulses from the Crab pulsar have amplitudes greater than 1000 times the average pulse height (Lundgren et al. 1995). High‐energy cosmic rays can cause transient radio signals at the surface of the Earth. The interaction of high‐energy particles from space and the Earth's atmosphere produces an "extensive air shower." Pair production in the shower creates populations of electrons and positrons that are systematically separated by the magnetic field of the Earth, setting up a current that produces a radio pulse (Kahn & Lerche 1966). These are only a few examples of known sources of transient astronomical radio emission.

Other sources have been postulated, but not observed, to produce short radio pulses. For example, Colgate, McKee, & Blevins (1972) and Colgate (1975) have predicted that a Type I supernova should radiate an electromagnetic pulse at radio frequencies: during the collapse, the expanding envelope of the white dwarf acts as a "conducting piston," compressing the transverse magnetic field, thus producing a short pulse of radio emission. Efforts to detect such pulses are described by Meikle & Colgate (1978) and Phinney & Taylor (1979). Another example is "exploding" black holes, predicted by Hawking (1974): the decrease in black hole mass due to quantum radiation causes an increase in surface gravity, which in turn increases the emission rate. Black holes near the ends of their lives would radiate intensely, releasing 10³⁰ ergs in the last 0.1 s. Rees (1977) has speculated that the expanding sphere of electrons and positrons would act like the conducting piston described by Colgate for supernovae, similarly producing a short radio pulse. Meikle (1977) and Phinney & Taylor (1979) have reported unsuccessful searches for these pulses.

Of particular interest for radio transient searches is the association between high‐energy emission and radio emission. Mattox (1994) reports searching the Compton Gamma‐Ray Observatory/EGRET phase 1 full‐sky survey for gamma‐ray emission from X‐ray–selected BL Lac objects and from the 200 brightest radio‐quiet quasars. None was detected, while EGRET did detect ∼40 radio‐loud quasars and radio‐selected BL Lac objects. This seems to suggest that "apparent gamma‐ray emission is intimately linked to apparent radio‐emission" (Mattox 1994), which is perhaps not surprising since the conditions that produce high‐energy emission (relativistic particles), in the presence of even a weak magnetic field, produce radio emission through the synchrotron mechanism. This association has been noticed by others as well (e.g., Paczyński & Rhoads 1993). The presence or absence of radio emission from high‐energy sources can yield clues about their workings.

Gamma‐ray bursts (GRBs) have been observed to produce emission at longer wavelengths following the gamma‐ray event (see van Paradijs, Kouveliotou, & Wijers 2000 for a review). These so‐called afterglows were first detected at radio wavelengths in the GRB of 1997 May 8 (Frail et al. 1997), and the field has since matured such that a catalog of radio afterglows is possible (Frail et al. 2003). The afterglows are believed to be due to emission from shocks produced when a relativistic fireball interacts with an ambient medium (Mészáros 2002 reviews models). Another possibility, which has not been detected, is a prompt radio burst associated with the GRB event itself. Again, energetic charged particles in a magnetic field might serve as a source; for example, Usov & Katz (2000) suggest that strong low‐frequency radio emission might be generated by time variability in the current sheath surrounding a magnetized jet. Examples of other ideas relevant to the detection of prompt radio emission from GRBs may be found in Palmer (1993), Hansen & Lyutikov (2001), and Sagiv & Waxman (2002).

Energetic particles can produce radio emission under other conditions as well. It was suggested some time ago (Askar'yan 1962, 1965) that high‐energy neutrinos and cosmic rays would generate coherent Cerenkov emission as they travel through a dense dielectric medium such as water, ice, the Earth, or the Moon. The physical process is similar to that of the extensive air shower discussed earlier, involving the production of a shower with an imbalance of charge. The spectrum of Cerenkov light from such an event would be very broad, including radio and optical emission. The generation of coherent radio bursts has been verified in accelerator experiments (Saltzberg et al. 2001). The pulses generated by particles traveling through the lunar regolith are expected to be very bright and very short: more than thousands of janskys for the highest energy particles, with durations on the order of a nanosecond (Alvarez‐Muniz & Zaz 2000). Hankins, Ekers, & O'Sullivan (1996) report an unsuccessful search for such emission.

Experiments to detect transient astronomical radiation at radio wavelengths have traditionally followed one of two approaches. When transient radio emission is thought to originate from a particular source or region of the sky, a high‐gain, small solid‐angle approach is used. A high‐gain radio telescope pointed at the region of interest provides good sensitivity and rejection of signals outside the region. When the location of the source of radiation is unknown, a low‐gain, large solid‐angle approach is more appropriate. Radio telescopes with large beams provide coverage of large fractions of the sky, but at the cost of reduced sensitivity, since the angular extent of any discrete source is likely to be much smaller than the telescope beam.

Some good examples of the high‐gain, small solid‐angle approach are those that were prompted in the early 1970s by Weber's reports of detections of pulses of gravitational radiation (e.g., Weber 1970). Since the gravitational waves were reported to have originated from the Galactic center, attempts to detect radio frequency activity focused on that region. Partridge & Wrixon (1972) monitored the Galactic center with two radiometers separated by 100 km. The first, at 16 GHz, provided a beamwidth of ∼12^', sensitivity of ∼100 Jy, and response time of 0.5 s. The other, at 19 GHz, provided a beamwidth of ∼12°, sensitivity of ∼10⁶ Jy, and response time of 3 s. Astronomical events were to be identified by their simultaneous appearance in the records at both sites. After 90 hours of monitoring, they detected no coincidences. Hughes & Retallack (1973) observed the Galactic center at 858 MHz with beamwidth 1 4, sensitivity 85 Jy, and response time 1 s. In 207 hours of monitoring the Galactic center, they report 97 detections of pulses. However, their interference rejection scheme is unclear, making uncertain their conclusion of the existence of discrete radio pulses from the Galactic center. O'Mongain & Weekes (1974) used the Mount Hopkins Observatory 10 m optical reflector (which was fitted with two outboard 4.6 m radio reflectors) to make a more general search including the Galactic center, the Coma cluster of galaxies, and the Andromeda galaxy at three radio frequencies and one optical frequency. In approximately 200 hours of observing, they detected no events of extraterrestrial origin, which were to be identified by the differences in pulse arrival times among frequencies, due to dispersion by the interstellar medium.

Other work has attempted to observe a much larger fraction of the sky with lower sensitivity. For example, Charman et al. (1970) assembled a system of five receiving stations in Great Britain and Ireland with station separations ranging from 110 to 500 km. Each station had similar receiving systems at 151 MHz, consisting of two half‐wave dipole antennas operating as phase‐switched interferometers, and receivers with sensitivities of ∼10⁵ Jy and response times of ∼1 s. Interference rejection was accomplished by the requirement of fivefold coincidence. After ∼2400 hours of observations, they detected no events. Mandolesi et al. (1977) used four receiving systems at Medicina (Bologna, Italy) operating at 151, 323.5, 330.5, and 408 MHz. On 1976 August 16 all four instruments detected a radio burst, while 60 s earlier, a gamma‐ray burst was detected by three satellites and one balloon‐borne detector. Strong rejection of local interference was provided by an independent observation made at 237 MHz at the Astronomical Observatory of Trieste (400 km distant from Medicina). The investigators estimated the probability of a random coincidence between the fivefold radio event and the gamma‐ray burst at 8 × 10^-5. Using geometric arguments, they localized the radio burst to a region on the sky. Unfortunately, later work on the gamma‐ray burst data from the balloon‐borne detector (Sommer & Müller 1978) produced a position for the gamma‐ray event that was well outside the radio emission error box, apparently ruling out a common origin for the gamma‐ray and radio events. But the possibility remains that the radio event was astronomical in origin. Hugenin & Moore (1974) used two receiving systems separated by several hundred kilometers to monitor the sky at 270 MHz. Each station consisted of a helical antenna with beam area of ∼1 sr centered on the north celestial pole and a multichannel receiver that provided a sensitivity of ∼10⁴ Jy (1 σ). The data were displayed with response times τ = 20 ms to 1 s on an oscilloscope and recorded by continuously photographing the oscilloscope screen. Interference was rejected by requiring coincidence between sites and by examining for the frequency dispersion expected in extraterrestrial signals due to the interstellar plasma. In 213 hours of observing, they detected no events. Amy, Large, & Vaughan (1989) constructed the Molonglo Observatory Transient Event Recorder (MOTER), which operated at 843 MHz in parallel with normal Molonglo Observatory Synthesis Telescope (MOST) synthesis observations. This system used 32 total‐power fan beams spaced at full beamwidth intervals of 44^''. When a transient event with flux density ≳10 mJy and duration 1 μs to 800 ms occurred in the field of view, MOTER compared the signals in the fan beams. Sources closer than about 3000 km were significantly out of focus, and thus appeared in many beams simultaneously, while signals appearing in one beam only were thought to be due to random noise. In this way MOTER was able to reject signals of local origin. A signal that was not rejected was localized to an arc segment on the sky corresponding to the fan beam in which its maximum appeared. Over a full synthesis observation, the resulting ensemble of arc segments intersected at a point, the location of the source. While MOTER lacked instantaneous large solid‐angle coverage, in time much of the southern sky was surveyed. In ∼4000 hours of observations, Amy et al. (1989) detected only previously known pulsars.

We describe here a large solid‐angle, low‐gain system for detecting transient astronomical radio emission at 611 MHz on timescales of a few minutes or less. Similar in spirit to the work of Charman et al. (1970) and Mandolesi et al. (1977), STARE updates previous efforts through the use of modern technology. The wide availability of fast computers and other hardware permits the collection of data in digital form. Such a data record makes possible a variety of analyses both during data collection and afterward, presenting a greater opportunity for discovery than the data sets from the 1970s that were generally in the form of analog chart records.

The STARE system consists of detectors at three geographically separated sites: one at the VLBA³ station in Hancock, New Hampshire, another at the VLBA station in North Liberty, Iowa, and the third at NRAO in Green Bank, West Virginia. The sites are close enough that they see the same part of the sky, while they are distant enough that radio frequency interference at one of the sites (from terrestrial or low‐altitude transmitters) will not be detected at the others. This provides a powerful filter for selecting signals of celestial origin: any signal that does not appear simultaneously at all three sites is rejected. These locations were chosen because they all have operating radio telescopes, and so are expected to have low levels of RFI, and because NRAO was kindly willing to support STARE operations. The STARE instruments are highly self‐sufficient, requiring only a standard AC electrical power connection, an Internet connection, space for the antennas and electronics, and a staff member willing to perform occasional minor crisis intervention. The apparatus at a site consists of several systems: an analog receiving system, a Global Positioning System receiver for providing accurate time, and a PC that collects data and controls everything at the site. The entire system is overseen by a workstation at MIT, which is also where the data are archived. Figure 1 shows the organization of the entire system.

Zoom In Zoom Out Reset image size

The analog receiving systems each consist of an antenna, a receiver, and a detector. The antennas are of the "crossed dipoles in a cavity" type, shown schematically in Figure 2. This style of antenna provides broad sky coverage, although at the expense of sensitivity. Laboratory measurements (by J. Barrett, P. McMahon, and W. Baumgartner) on a scale model of the antennas found a broad beam with effective solid angle 1.4 sr, yielding a sensitivity of 6.1 × 10^-5 K Jy⁻¹. The receivers are dual‐channel superheterodyne total‐power radiometers and are each divided between a front end that is outdoors attached to the antenna and a back end that is indoors. The front end begins with low‐noise ambient temperature preamplifiers (Harris & Lakatosh 1987) with effective noise temperatures in the range 100–150 K. The front end also includes a laboratory‐calibrated noise source for gain calibration and system temperature measurements. The back end performs bandpass filtering, further amplification, frequency downconversion, square‐law detection, and antialias filtering at 25 kHz. The analog receiving systems were designed to provide sensitivity to the two orthogonal circular polarizations, with beam patterns independent of source azimuth. However, on account of an error in the construction of the feeds, they receive two orthogonal elliptical polarizations, which vary with source azimuth. We avoid this problem by summing the two received polarizations. It can be shown (Katz 1997), and is evident from symmetry, that the orthogonality of the received elliptical polarizations means that their sum is independent of the azimuth of the source. Thus in practice we are able to calibrate only the sum of the total power in both polarizations.

Zoom In Zoom Out Reset image size

In order to reject local interference, we must be able to determine simultaneity from site to site, requiring an accurate timekeeping method. This is provided by the Global Positioning System (GPS). At each STARE site is a GPS antenna and receiver. When initially installed, each GPS system was allowed to compute its position continuously for several days. The data were then suitably averaged to increase the accuracy of the position determinations. Those positions were then programmed into the GPS systems, allowing four satellites to be used for timekeeping alone. In this "static timing mode," the GPS receivers have a claimed timing accuracy of ±100 ns.

Data acquisition and all other activity at each site are controlled by a standard desktop PC. The filtered square‐law detector outputs are digitized at 50 kHz by a 12 bit analog‐to‐digital converter card which is clocked by a 100 kHz signal from the GPS receiver. Time stamps are assigned to the data by switching in a short pulse from the GPS receiver and finding the sample during which the pulse is detected, yielding a time stamp accuracy of 20 μs. The samples are boxcar integrated, then collected and saved on the disk for some specified period, then are transferred to a workstation at MIT. In addition to performing overall coordination and data archiving, the workstation is responsible for the data analysis: the data files from each site are analyzed individually to produce a record of transient radio signals there. The site event records are then compared to find three‐way coincidences.

The entire STARE system runs automatically, producing a daily report of site events and coincidence detections. The computer systems are set up so that most routine maintenance can be performed without traveling to the sites. The system is quite robust, having required human attention on average less than once a month. Sometimes attention is necessary to clear a fault condition, but most often routine maintenance (e.g., clearing a full disk) is the cause. Occasionally a phone call to a site is necessary (e.g., to identify and replace an electrical fuse blown during an electrical storm).

In normal operation, the power received by each antenna at 611 MHz in 4 MHz bandwidth and two polarizations is boxcar averaged for 0.125 s and recorded. Every 15 minutes, the noise source in the front end is switched on for 2 s, to provide receiver gain calibration. Each hour, the data from the previous hour are uploaded to the workstation at MIT, where they are processed and archived.

Coincidences among sites are detected by the simplest possible method: the records of each site are examined to find events at that location, then the lists of single‐site events are compared to find instances where all three sites recorded events simultaneously. Note that with this simple scheme, the overall sensitivity to three‐way coincidences is determined by the site with the poorest detection sensitivity. If one site fails to detect an event, the event cannot be identified as a three‐way coincidence. More sophisticated schemes could no doubt produce better detection sensitivity.

The first step, the detection of events at each site individually, is performed using a two‐pass sliding‐window baseline fit. On the first pass, a quadratic model is fitted to the data in a 240 s window using a least‐squares algorithm, and the dispersion σ of the data around this fit is calculated. On the second pass, the model is again fitted to the data in the 240 s window, this time using the robust estimation method described by Press et al. (1992, § 15.7). We implemented the method with a Lorentzian weighting distribution with the width set to the dispersion σ calculated in the first pass. This method was chosen so that we could follow the wandering baseline in the data without our fit being skewed by outlier points, of which there are many, including the transient signals we are trying to detect. Note that this results in a loss of sensitivity to events longer than a few minutes. Greater sensitivity to long‐lived events could be achieved by increasing the width of this boxcar averaging window, at the expense of reducing the baseline‐removal effectiveness for shorter time periods.

Once the baseline fit is determined, the data are examined for samples that deviate from the baseline by more than some threshold specified as a multiple of σ. When a sample is seen to deviate by more than the threshold, an event is considered to be in progress, and the next sample is examined. This continues until the deviation falls below one‐half of the threshold, at which time the event is considered to have ended. We use this adaptive threshold to reduce the incidence of long, temporally spiky events being broken up into multiple events as the flux density level varies around the threshold. Once the events are detected, they are classified as "single point" or "multipoint," depending on whether the thresholds were exceeded only in a single sample (i.e., event duration ≤0.125 s) or in more than one consecutive sample.

The thresholds for event detection were chosen so that from site to site, the flux density required to trigger an event is about the same. Using receiver noise temperature 150 K, zenith antenna sensitivity 6.1 × 10^-5 K Jy⁻¹, integration time 0.125 s, and bandwidth 4 MHz, we calculate a theoretical zenith flux density sensitivity for our systems of about 4 kJy. Of course, in practice the system temperatures are higher than the receiver temperatures because our 1.4 sr beam admits RFI from sources far and wide, degrading the system sensitivity. We found that the system temperatures also vary widely with time. At Green Bank and North Liberty, the median system temperatures through 18 months of observations were similar at about 200 K. At Hancock, the system temperatures were typically higher by a factor of 2. To make the zenith flux density sensitivities consistent from site to site, we chose a threshold of 5 σ for transient detection at Green Bank and North Liberty and 2.5 σ at Hancock. These thresholds correspond to a zenith flux density triggering threshold of about 26 kJy. For sources away from zenith, the threshold is of course higher due to the decrease in antenna gain.

Once the single‐site event lists are produced, they are compared to find events detected simultaneously at all three sites. Three‐way coincidences are classified as "single point" if the three single‐site events are all single‐point events, "multipoint" if they are all multipoint events, and "mixed type" if the simultaneous single‐site events are a combination of single point and multipoint. To assess the coincidence detection sensitivity of STARE, we examined the flux density sensitivity at each site for every minute of time that all three sites were on‐line. We then chose the poorest sensitivity of the three as the STARE coincidence sensitivity, since the site with the poorest sensitivity would determine the overall sensitivity. The results are shown in a normalized histogram in Figure 3. Not shown are the 5.6% of the values that lie beyond the right side of the plot, due to noisy times at one or more of the sites. The median overall STARE zenith sensitivity of 26.7 kJy is very close to the median single‐site detection sensitivity of 26 kJy, indicating that most of the time, the three sites run with comparable sensitivities.

Zoom In Zoom Out Reset image size

Event data were calibrated in several steps. First, the receiver gains were determined using the noise source on/off data. These were used to convert the data to antenna temperatures in each polarization channel. The antenna temperatures were summed to find the total antenna temperature and to avoid the problem with the unknown polarizations of the feeds (see the discussion of the analog receiving systems in § 2.1). Then, using the antenna sensitivity determined from measurements on a scale model of the antennas, the antenna temperatures were converted to the flux densities that would be measured if the source were at the zenith. This resulted in a lower limit on the flux density of the source. For sources of known elevation, such as the Sun, the zenith flux densities were corrected for the elevation response of the antenna, yielding a source flux density measurement.

The calibration is rather coarse. In practice we find that the flux densities of solar radio bursts measured by the STARE systems generally agree from site to site only within a factor of 2 or so. Comparing measurements between sites, we find that the Green Bank instrument systematically reports flux densities approximately a factor of 2 larger than those reported by Hancock, which in turn are approximately a factor of 2 larger than those reported by North Liberty. In addition, there is significant variation of these ratios from event to event. We believe this is due primarily to uncertainties in two parts of the receiving systems. First, the beam patterns of the antennas were determined on a 1/10 scale model and may not transfer very well to the actual feeds. This would cause elevation‐dependent variations in the calibration and could account for the scatter in the systematic differences between sites. Second, the excess noise temperatures of the noise sources used for gain calibration were measured when the systems were constructed and may have changed over the several years of aging they have experienced. This could be the cause of the overall systematic differences between sites.

Clearly the calibration accuracy could be improved by measuring or computing numerically the antenna beam patterns and by measuring the present noise source characteristics. But for our purpose here, we consider the calibration accuracy of a factor of a few to be adequate and defer the improvements to later work.

In 18 months of operation, STARE detected hundreds of thousands of events at Green Bank and North Liberty and millions at Hancock with its weaker triggering threshold. The exact numbers are given in Table 1.

Using the mean single‐point event rates computed from the single‐site event detection results, we can estimate the rate of accidental single‐point coincidences among the sites. For a mean event rate r, the probability that a time interval δt contains an event is rδt (assuming r≪1/δt, which is true for STARE since δt = 0.125 s). Thus for three sites with mean event rates r₁, r₂, and r₃, the probability that a time interval δt contains an event at all three sites is r₁r₂r₃(δt)³. The mean time ΔT between accidental three‐site coincidences is then ΔT = 1/r₁r₂r₃(δt)². Using the rates given in Table 1, we find ΔT ≈ 3 years. With only two sites (even those with the lowest event rates, Green Bank and North Liberty), this time is ΔT = 1/r₁r₂δt ≈ 11 days. This illustrates the power of the coincidence requirement in filtering out local radio frequency interference. It also makes clear that the third site is required to reduce the accidental coincidence rate to a manageable level.

A similar examination of accidental coincidence rates for multipoint events is not necessary, since their temporal structure provides much more information than is available with single‐point coincidences. Accidental coincidences may be ruled out simply by direct comparison of the time‐series data from each of the sites. Unless the events have the same shape, they can be rejected as events of interest.

In approximately 18 months of data collection, the three STARE sites recorded well over 4 million radio bursts. Of these, 3898 were identified to be in temporal coincidence among all three sites: 1859 at Hancock, 1069 at Green Bank, and 970 at North Liberty. The numbers are unequal because of the spiky temporal nature of many of the events. Despite the use of the adaptive threshold algorithm described in § 2.2, the event detection algorithm tended to break up long events into multiple events, depending on exactly how the measured power varied around the detection thresholds. In many instances, a single long event observed at one site was broken up into multiple events at the others, due to sensitivity differences between sites. In this case, STARE reported multiple coincidences despite one site reporting only a single event. To account for this effect, the data for each reported coincidence were examined visually to determine which of these multiple events were really just parts of the same overall event. After this reduction step, we found that the reported coincidences collapsed into 126 distinct events. Examining each of these more carefully, we found that 27 were accidental. Accidental events were identified using two criteria: the time dependence of the events differed obviously among the sites, and/or one or more of the sites showed a temporarily very high event rate, due to some local RF emitting phenomenon. Ignoring the accidental events, we find that STARE detected 99 events that appear to be of astronomical origin.

Since we expect that the Sun is the most intense source of transient radio emission in the sky, we compared the STARE events with those detected independently by a solar monitoring station. The US Air Force operates a worldwide Radio Solar Telescope Network (RSTN) for the purpose of producing warnings about solar weather events that could disrupt terrestrial systems. Conveniently for us, one of the stations of the RSTN is Sagamore Hill Solar Radio Observatory in Hamilton, Massachusetts, approximately 80 km distant from our STARE instrument in Hancock, New Hampshire. The RSTN monitors the Sun for transient activity at eight fixed frequencies, one of which is 610 MHz, the same as the STARE observing frequency. Thus the record from the Sagamore Hill RSTN station is useful to us for comparison and verification purposes.⁴ The RSTN 610 MHz system operates on an 8.5 m dish, using a dipole feed to measure a single linear polarization, recording the solar flux density once per second (Heineman & Ambrisco 1997).

During the 18 months of STARE three‐site data collection, the RSTN Sagamore Hill station reported 146 solar radio bursts at 610 MHz. After combining these events with the 99 detected by STARE during this time, we divided the ensemble of events into three groups: those detected by STARE only (58 events), those detected by RSTN only (105 events), and those detected by both STARE and RSTN (41 events). Of course, the first group is potentially the most interesting as it may include transient astronomical radio emission from nonsolar sources. The other groups are useful in understanding the behavior of the STARE system. To illustrate a typical event, we show in Figure 4 the event detected by STARE and RSTN on 1999 May 23 at 17:30 UTC.

Zoom In Zoom Out Reset image size

We first compared the peak flux densities of the events measured by STARE and RSTN, and we found that the Hancock site produces flux density estimates that are on average closest to those of the RSTN. The Green Bank site produces higher values, and the North Liberty site produces lower values, more or less consistent with the systematic flux density scale differences described in § 2.3, although with a large scatter around these averages. Note however that we do not expect particularly good agreement between STARE and RSTN flux density measurements since STARE measures the total flux density in both polarizations, while RSTN measures a single linear polarization. The exact ratios of the measurements depend on the polarization of the received radiation.

We used the full ensemble of 204 events to help characterize the triggering criteria of STARE and RSTN. Figure 5 shows plots of event flux density against event duration for both RSTN‐measured values and STARE‐measured values (from Hancock, since it is in closest agreement with RSTN). We can draw several conclusions from these plots. From the left plot, we see that STARE is more sensitive to short‐duration events. This is to be expected, since STARE takes eight samples per second, while the RSTN takes one. From the right plot, we see that the RSTN has better sensitivity. This is also to be expected, since RSTN uses a parabolic dish that tracks the Sun. To properly interpret the right plot in Figure 5, note that RSTN reports only the start and end times for events, with 1 minute accuracy, accounting for the distinct columns of points in the plot. For this plot, we arbitrarily assigned a duration of 1 s to events listed as starting and ending in the same minute.

Zoom In Zoom Out Reset image size

We then examined the 105 events detected by RSTN only and found three reasons that they were not detected by STARE. A subgroup of 74 events was simply too faint for detection by STARE. For another 25 of the events, manual examination of the STARE data showed that the events were indeed recorded but that they did not exceed the trigger threshold at all three sites, so they were not identified as three‐way coincidences. For the remaining six events, there were no data recorded by one or more STARE sites at the time of the RSTN event, on account of the systems being off the air for maintenance or other purposes. From these results we gain confidence that during the 18 months of three‐site data collection, STARE detected the events we expected it to detect.

Finally, we examined the final group of 58 events: those detected by STARE but not by the RSTN. From the left plot in Figure 5, we see that many of these events are of short duration, suggesting that they might have been of solar origin but were too short to trigger a radio burst alert from RSTN. In addition, all of the events occurred during daytime hours. To determine unequivocally whether these events were from the Sun, we obtained the RSTN 1 s data stream and compared it directly with the STARE records. For 51 of the events, the signal was easily discernible in the Sagamore Hill RSTN record, matching the STARE event well in time and shape. A somewhat deeper investigation into why these events were not reported by RSTN determined that when events are detected by RSTN, a human examines the data to decide whether burst alerts should be issued (Heineman & Ambrisco 1997), so we should not count on the burst alerts as a complete record. The other seven events were not found in the Sagamore Hill RSTN record because of gaps in the data, due to equipment outages or calibration. For these, we obtained and examined the data from two other RSTN sites: San Vito, Italy, and Palehua, Hawaii. Five of the remaining events were unambiguously identified in these records. The two remaining events occurred at times when no RSTN data were available; at San Vito, the Sun had already set, and at Palehua, the events occurred during gaps in the data. However, we suspect that these two are due to solar radio bursts as well. For one event, the Sagamore Hill data resume several seconds after the STARE event time and show the final moments of an event in progress. For the other, a large solar radio burst (detected by both STARE and RSTN) occurs less than 30 minutes after the event. In both the RSTN and STARE records we see that events tend to be clustered in time, so it would not be unusual if this event were related to the following large burst. In addition, both of the unidentified STARE events are very short (<0.5 s) and so are unlikely to have been detected by RSTN, as discussed above in reference to Figure 5. Although we cannot definitively associate these two events with solar radio bursts, we believe that the indirect evidence indicates that such an association is warranted. From these results we deduce that all of the astronomical signals detected by STARE were due to solar radio bursts.

We thank the many people who have made STARE possible. At the National Radio Astronomy Observatory, M. Goss offered the use of VLBA stations for STARE receivers, while F. J. Lockman offered the use of space at Green Bank. Development and operation of STARE would have been impossible without the help of T. Baldwin and D. Whiton at the Hancock VLBA station, D. J. Beard and B. Geiger at the North Liberty VLBA station, and M. McKinnon and R. Creager at Green Bank. Some initial STARE testing was performed at Hat Creek Radio Observatory, where space and support were offered by D. Backer, and the work was aided by R. Forster and M. Warnock. Development of the STARE instrumentation was helped along by advice from A. Rogers, L. Beno, and P. Napier. J. Ellithorpe contributed to the computer work, and J. Barrett provided many hints and lessons. H. Coffey and E. Irwin at the National Geophysical Data Center provided the data from the RSTN telescopes, and TSgt C. Hoffman, USAF, of the Sagamore Hill RSTN station answered many questions about RSTN operations and methods. Finally, we thank the anonymous referee whose insightful comments led to significant improvements in this article.

We are grateful for the support STARE has received from a variety of sources, including a David and Lucille Packard Fellowship in Science and Engineering, a Bruno Rossi Graduate Fellowship in Astrophysics, the MIT Class of 1948, a National Science Foundation Presidential Young Investigator grant (AST 91‐58076), and a Postdoctoral Fellowship from the Smithsonian Institution.