Massachusetts Institute of Technology Department of Physics and Center for Space Research Cambridge, MA 02139-4307
Keywords: Optical Transients, Instrumentation, Automatic Telescopes, Gamma-ray Bursts
1. Introduction
The quest to understand the source of gamma-ray bursts (GRBs) began with their first detection in the late 1960's (Klebesadel, Strong, and Olson 1973). However, despite over twenty years of research, no definitive picture of the GRB source has emerged. In fact, first results from the Burst and Transient Source Experiment on the Compton Gamma Ray Observatory, which many had hoped would confirm models involving a galactic population of neutron stars as the source of GRBs, have led to increased confusion about the nature of GRB sources. There is, at present, no consensus regarding even the most fundamental characteristics of GRBs: estimates of their distances from the Earth and intrinsic luminosities range over many orders of magnitude.
One major reason for the uncertainty about GRB sources is the fact that they have only been detected during outburst: there is not a single confirmed detection of a quiescent GRB source in any waveband. Follow-up observations of GRB error regions at radio, IR, optical, and x-ray wavelengths have revealed many candidate sources, but have failed to unquestionably link a candidate source with a GRB (Radio: Hjellming and Ewald 1981. IR: Apparao and Allen 1982; Schaefer 1989. Optical: Chevalier, et al. 1981; Fishman, Duthie, and Dufour 1981; Laros, et al. 1981; Pedersen, et al. 1983; Schaefer, Seitzer, and Bradt 1983; Schaefer and Ricker, 1983. X-ray: Helfand and Long 1979; Pizzichini, et al. 1985). This failure is due in large part to the size of the error regions of GRBs, which is a result of the poor intrinsic angular resolution of gamma-ray burst detectors. Most GRBs error regions are large, with areas measuring hundreds of square arc-minutes to square degrees: of the hundreds of GRBs detected to date, only a handful have error regions small enough (< ~ 10 sq. arcmin) to allow reasonably deep searches for a quiescent counterpart at other wavelengths (Laros, et al. 1981; Chevalier, et al. 1981; Pedersen et al. 1984).
Thus, it has become clear that the study of GRBs would benefit greatly from a high-precision localization of a GRB. In the early 1980s, several experiments were initiated to search for contemporaneous optical emission from GRBs (Schaefer 1981; Schaefer, Vanderspek, Bradt, and Ricker 1984; Hurley, private communication), in the hope of pinpointing the celestial coordinates of the bursts using optical detectors with high angular resolution. Experimenters were encouraged by the discoveries of Schaefer (1981) and Schaefer, et al. (1984), which demonstrated that GRBs may have optical counterparts, and that these counterparts might be quite bright: as bright as V ~ 3 for a one-second optical burst. These early experiments met with limited success, however, due in part to the rareness of GRBs and the time-intensive data analysis methods required. In 1983, the development of the Explosive Transient Camera (ETC) began at the Massachusetts Institute of Technology (Ricker, et al. 1984). This instrument was designed to detect short-timescale (t ~ 1s) optical transients of V < ~10 and localize the source of the emission to ~15". The ETC was designed to be completely automatic, able to perform observations and data reduction without the need of an observer on site. The ETC was to be deployed as a ground-based optical counterpart to space-borne GRB detectors, in the hope of detecting any optical light emitted contemporaneously with the GRB, and thereby improving the localization of the GRB by many orders of magnitude.
The ETC is a wide-field sky monitor which can detect transients as faint as V ~ 10. The ETC consists a bank of wide-field CCD cameras, each sensitive to optical transients with risetimes of the order of one second and capable of localizing bright transients to within ~20". The ETC conducts a program of monitoring the night sky for optical transients automatically each night.
The ETC was built and tested during the mid-1980s, and is presently set up at a dedicated site at the Kitt Peak National Observatory in Arizona. The ETC has been operating automatically in January, 1991.
This paper is organized as follows. Section 2 is an overview of the ETC hardware and operation. Section 3 provides a summary of the operational characteristics of the ETC, such as limiting sensitivity and observing efficiency. Section 4 reviews the scientific goals of the ETC. Section 5 is a detailed description of the ETC hardware, while Section 6 describes the ETC's operational software. Section 7 lists future modifications to the ETC intended to increase its field-of-view and operational efficiency.
2. Overview of the Explosive Transient Camera
Each ETC camera consists of a cooled CCD in the focal plane of a commercial 24mm-format SLR lens. The field-of-view of each CCD camera is 20.5 x 15.3 degrees, consisting of 390 x 292 pixels, each subtending 2.2 arc-minutes (a typical ETC image is shown in Figure 1).
Figure 1: A full image from an ETC camera: east is to the right, north is at the top. The field-of-view of this image measures 20.5 x 15.3 degrees.
The cameras are mounted in two sets of eight on two telescope drives which track the sky during observations. The declinations of the cameras are fixed, but the two telescope drives are independent, and can steer each set of eight fields-of-view to any hour angle. During observations, the fields-of-view are oriented in a manner which minimizes the zenith distance of the cameras: a typical arrangement of fields-of-view is shown in Figure 2.
Figure 2: The layout of the ETC CCD camera fields-of-view on the night sky. The view shown is a zenith projection from Kitt Peak: the outer circle represents the horizon and the inner circle an altitude of 30 degrees (2 airmasses). The declinations of the centers of the cameras are roughly 4 degrees, 16 degrees, 28 degrees, and 40 degrees. The field-of-view of each CCD camera measures 14 degrees x 11 degrees.
The ETC cameras are housed in a dedicated building on the summit ridge of Kitt Peak. The building is covered with a fully-retractable roof, which allows the ETC cameras access to the full sky above 30 degrees altitude. A summary of ETC CCD camera characteristics is given in Table 1.
The ETC has been designed to be completely automatic, in order to reduce the trouble and cost of constant human attention to the instrument. The ETC computers have complete control of all of the ETC instrumentation, and so are able to open and close the roof, slew the telescope drives, and sense weather conditions (precipitation, windspeed and temperature). The ETC software determines when and how to operate, and makes decisions regarding operating conditions in the manner that an observer on site would. In short, no human observer is required at the site, because the ETC computer system performs all the functions that the observer on site would.
The method used by the ETC to detect optical transients is straightforward: compare consecutive CCD images in a search for 'new' stars or 'suddenly brighter' stars. When observing, the ETC's CCD cameras simultaneously take a continous series of contiguous, precisely-timed exposures of the night sky. The image data from each CCD camera is analyzed by its dedicated microprocessor, which compares consecutive images for sudden, point-like brightenings. Any detected brightenings are reported to the central control computer, which tabulates and correlates flash reports from all cameras. If two cameras monitoring the same field-of-view detect a brightening at the same celestial coordinates, the brightening is considered a bona fide celestial optical transient, and data from the event is stored. If no events are detected, no action is taken; in either case, the reduction of CCD image data stays on its precisely-timed schedule. Data taken by the ETC, which consist of a time series of images of the flash and of several SAO stars (as astrometric and photometric references), is stored on hard disk and later transferred to MIT via Internet.
3. Operational Characteristics
In this section, we discuss those characteristics of the ETC which describe the quality of observations performed by the ETC. These characteristics are 1) the sensitivity of the ETC cameras to the detection of field stars, 2) the sensitivity of the ETC cameras to the detection of optical transients, and 3) the observing efficiency of the ETC.
where A is the area of the lens (in cm2), texp is the exposure time, Dl1000 is the bandwidth of the filter in units of 1000, eatm, efilter, elens, and ewindow are the transmission efficiencies of various optical elements in the system, and eCCD is the efficiency of the CCD (in electrons/photon). (A more precise method would involve integrating the stellar spectrum over the bandpass, taking into account the variation in throughput with wavelength of the various elements of the system).
The significance of the detection is measured by its signal-to-noise ratio,
where Np is the number of pixels subtended by the stellar image, and
is the total variance per pixel in the CCD image, where sR is the CCD readout noise (in electrons).
At present, each ETC camera is equipped with a 24 mm, f/1.4 Nikon lens and a filter which passes most B-,V-, and R-band radiation. The measured values of the parameters in equation (1) are A = 4.9 cm2, texp = 5.0 seconds, Dl1000 = 3, eatm = 0.8 (Allen 1976), efilter = 0.75, elens = 0.9, ewindow = 0.92, eCCD = 0.3 e-/photon (averaged over the entire passband). In addition, the value of sR is typically 20 e-, and the sky rate during dark time has been measured as 300 e-/s (corresponding to ~20.5 mag/sq. arcsec over the entire passband, which is consistent with published Kitt Peak values). With these values, equation (1) reduces to ne = 11.0 fph. Assuming that a 15th magnitude star gives fph = 1 ph/cm2/sec/1000, this reduces further to m = 17.6 - 2.5*log10(ne). The limiting magnitude is given by solving equation (2): for Np=9 and S/N=4, the limiting value of ne is 530 electrons, which reduces to a limiting magnitude of 10.8. As a comparison, a histogram plotting the frequency of detection of a star of a given magnitude vs. magnitude for a typical ETC image is shown in Figure 3.
Figure 3: A plot of the frequency of detection of stars of a given magnitude vs. magnitude. The data were taken from a five-second exposure made in May, 1991. Because of the large-bandpass filter used on the ETC, the magnitude corresponds to a "B+V" magnitude, spanning the wavelength range of ~4000 to ~6500: the magnitudes were calibrated with known A-stars in the field-of-view. The level of significance of the detections was set at S/N = 4.
The result is that a transient of fOT photons/cm2/sec/1000 of duration tOT is detected at its peak as ne electrons, where
Equations (2) and (3) can be solved for the limiting sensitivity in the same manner as equations (1) and (2) were, above; the only differences are 1) a burst duration must be assumed, 2) the value of Np in equation (2) should be set to 1, and 3) the term sT2 should be multiplied by two to account for the fact that two pixels are being compared. Using the values given in the analysis above, equation (3) reduces to ne = 2.2fOTtOTesplit, and the magnitude of a burst is calculated as mOT = 15.9 - 2.5*log10(ne/(tOT*esplit)). If the detection threshold is set at S/N = 10, then the detection threshold is ne = 490 electrons, and the detection threshold is given by
Equation (4) demonstrates the strong dependence of the calculated detection threshold on tOT and esplit. If we assume an intermediate value of esplit = 0.5 and a transient duration of 1.0 seconds, the limiting transient magnitude is calculated to be mlim = 8.4. If, on the other hand, esplit = 0.8 and tOT = 5 seconds, then mlim = 10.6. In Figure 4, the effect of the random nature of the location of the centroid of an image on the possible values of esplit is shown.
Figure 4: Because the ETC is sensitive only to the variations in brightness of individual pixels, the probability of detecting a transient depends not on the integrated brightness of the transient, but rather the brightness of the peak pixel within the image of the transient. Thus, the probability of detection of a transient depends on the FWHM of the point-spread function (PSF) of the image: as the FWHM of the image increases, the fraction of the total light which is in the brightest pixel of the image decreases. In this Figure, we demonstrate this effect with images of four different FWHM: plotted is the fraction of the total light in the brightest pixel, calculated as a difference in sensitivity in magnitudes. The spread in each plot is due to the fact that the fraction of total light in the brightest pixel is a function of the location of the image centroid within the brightest pixel.
The probability of seeing a particular value of esplit is given as a function of the FWHM (in pixels) of the seeing disk of the system: for the ETC, the typical FWHM is ~1 pixel. The effect of this distribution of esplit is to introduce a probability distribution in the limiting magnitude. The time of the onset of the optical transient with respect to the contiguous series of ETC exposures also plays a role in whether the transient is detected. Equation (4) implicitly assumes that, if tOT < texp, that the burst occurs entirely within one ETC exposure, and if tOT > texp, that the burst begins at the beginning of an ETC exposure. In the former case, there is, clearly, a probability that the burst will straddle two consecutive exposures, thus reducing the probability of its detection; in the latter case, the burst will start anywhere within an exposure, thus changing the probability of its detection. In Figure 5, the effect of burst duration and phase with respect to the ETC integration cycle is shown.
Figure 5: The time of the onset of an optical transient with respect to the contiguous cycle of ETC exposures can affect the probability of detecting the transient if the transient fluence is close to the fluence required to be detected by the ETC. A transient with fluence equal to the threshold fluence and duration, tOT, equal to the exposure time, texp, will rarely be detected, because the probability is high that the transient will straddle the boundary between two consecutive exposures; if this transient has tOT = 0.5texp, however, the probability of detecting the transient will rise to 50%. This effect is demonstrated in this Figure using sample values of tOT/texp.
Since the phase of the onset of the optical transient is random, Figure 5 gives the probability of the detection of a burst as a function of the strength of the burst (compared to the detection criterion). The probability of detecting a burst is unity for bursts well above the detection threshold: only near the detection threshold does the probability decrease.
The ETC is also sensitive to sudden brightenings of field stars. It is possible that certain classes of flare stars can brighten enough in a few seconds to be detected by the ETC. In addition, Schaefer (1989) postulates the existence of flare phenomena on "normal" stars. The requirements for detection of a field-star brightening are that the brightening be large enough and fast enough to trigger the ETC's algorithms. A star which brightens by ten magnitudes over one day may not be detected by the ETC because its rate of rise is too slow, while a star brightening 0.1 magnitudes in five seconds may not be seen because its magnitude increase is too small. Figure 6 plots the decrease in magnitude during one ETC exposure required for a field star brightening to be detected, as a function of the star's quiescent magnitude.
Figure 6: A plot of the magnitude decrease required of field stars in order to be detected by the ETC, as a function of the quiescent magnitude. The detection criterion is S/N = 10.
The value of eNIGHT is just the average over the year of the fraction of each day which is after evening astronomical twilight and before morning astronomical twilight, and is 0.37. The effect of the moon is taken into account in eDARK: as the ETC does not observe when the moon is within 1.5 hours of the meridian, the value of eDARK is 0.875. The value of eCLEAR is more subjective: it is the fraction of "useable" dark hours per year. The definition of "useable" varies from one astronomer to the next, but we estimate, based on our experience with the ETC on Kitt Peak, that eCLEAR is ~0.65. This value takes into account the fact that most of the months of July and August are lost to the summer monsoon season in southern Arizona. This value compares well with the value calculated by Kitt Peak staff of ~0.7 (Kitt Peak Newsletter, No. 28, 1991). Finally, the value of eOBSERVATION, which is the efficiency with which the ETC conducts observations on clear, dark nights, has been seen to be ~0.65 for the present ETC configuration. Using these values, the value of eETC is calculated to be 0.14: this implies that the ETC will conduct observations for 1200 hours each year, and that the total sky survey coverage will be over 400 steradian-hours per year.
4. Scientific Goals of the ETC
The primary goal of the Explosive Transient Camera is to detect an optical transient which is spatially and temporally coincident with a gamma-ray burster. However, because the ETC is sensitive to all sources of short-timescale optical transients, it will conduct a systematic survey of the night sky for optical transients with risetimes of the order of one second and peak magnitude V < ~10. As such a survey has not yet been performed, the prospect of serendipidous discoveries of new classes of astronomical objects by the ETC is very promising. The ETC is also sensitive to transient behavior from field stars, and so will be able to detect new classes of fast flare stars, or brightenings from "normal stars" (Schaefer, 1989). In the course of normal operations, the ETC will also create an enormous archive of wide-field optical images of the night sky to a limiting magnitude of mV ~ 11: this archive will be searched for transients with longer timescales (hours, days, weeks) and also serve as an image archive for other workers in the field.
One can make a rough estimate of the rate at which the ETC will detect optical counterparts to GRBs, given the ETC sensitivity, the log N(>S) - log S relation for GRBs, and an estimate of the ratio of gamma-ray to optical fluence, Sg/Sopt, for such GRBs. The number of bursts the ETC will detect in one year, NETC, is given by
where NGRBS (S > SETC) is the number of GRBs corresponding to the ETC fluence limit, SETC, WETC is the solid angle covered by the ETC, eETC is the fraction of a year that the ETC is actually conducting observations, and eGRB/OT is the fraction of GRBs which create optical transients. Assuming for now that eGRB/OT is unity, the ETC detection limit of V ~10 corresponds to SETC(optical) = 3.6 x 10-10 erg/cm2, assuming a one-second burst duration. The canonical value for Sg/Sopt is ~103 (Grindlay 1983; Schaefer 1981); thus, (Sg/Sopt) SETC(optical) = 3.6 x 10-7 erg/cm2. From Jennings (1982), NGRBS (S > 3.6 x 10-7) = ~1000. The value of WETC/4p is 0.03. The value of eETC was shown to be 0.14 in Section 3.2. Thus, NETC ~ 4 optical counterparts per year. If the ~ 3x higher GRB rates from BATSE are adopted (reference), NETC becomes of order ten.
The probability that the fields-of-view of BATSE and the ETC will overlap, eBATSE/ETC, is given by
where WBATSE and WETC are the solid angles covered by BATSE and the ETC, respectively, and eBATSE and eETC are the observing efficiencies of BATSE and the ETC, respectively. The values of eETC and WETC are given above; WBATSE/4p ~ .75 because of Earth occultation, and eBATSE = 0.85 (GRO Newsletter, Vol. 1, No.3). Thus, eBATSE/ETC is 0.003. Given the present rate of detection of bursts by BATSE of 92 in 109 days (GRO Newsletter, Vol. 1, No.3), BATSE and the ETC will overlap during a GRB roughly once a year.
Figure 7: A block diagram of the ETC computer and peripherals system. Each set of CCD control electronics controls eight CCD cameras, and each Trigger Processor Enclosure contains eight Trigger Processors, each corresponding to one CCD camera.
The OC is built around a Pacific Microcomputers 68000 CPU board running the Unix V7 operating system. Three Winchester hard disks provide 100 Mbytes of storage space for ETC software and data. An octal serial port card provides eight RS-232 serial ports. Data from the TPs enters the OC through a Heurikon HK68B/ME single-board computer. The CCD image data can be displayed on an AED color graphics unit. The OC is equipped with an Ethernet card to allow access to the Internet on Kitt Peak: when the network is unavailable, data and software in the OC can be accessed efficiently from MIT via a Telebit high-speed (>10 kbaud) modem. The OC synchronizes its system clock with WWV over a serial interface to a commercial WWV receiver.
The CCD control electronics are the Overseer Computer's interface to the CCD cameras. A block diagram of the CCD control electronics is shown in Figure 8.
Figure 8: A block diagram of the CCD control electronics. A single COSMAC system controls the clocking of eight CCDs. The clocking voltages for each CCD are supplied by dedicated digital-to-analog converters; the clock timing for eight CCDs is supplied by a single programmable sequencer.
On command from the OC, each set of CCD control electronics generates the clocking signals to read out eight CCDs. The video signals from all CCD are then amplified and digitized in parallel, and the digital data are transmitted to dedicated TPs. The TPs then analyze the image data in real time.
Each TP is a Heurikon HK68B/ME 68000-based single-board computer with 1 Mbyte of memory and DMA for efficient data reception and transmission. Data passes to the TP over a 250 kpixel/second parallel link from the CCD control electronics. The TPs are housed in groups of eight in a single Multibus enclosure. Communication between each of the eight TPs in an enclosure to the OC occurs via another Heurikon HK68B/ME called the Intermediate Level Processor (ILP). The ILP acts as a sophisticated post office for the eight TPs: it allows for more efficient operations by reducing the number of communications paths from the OC to the TPs from eight to one.
The peripheral hardware controller is the ETC's interface to much of the hardware in the system. Through the peripheral hardware controller, the OC controls the motion of the roof, and the motion of the sidereal drives; it also decodes the signals from the various weather sensors. A block diagrams of the peripheral hardware controller is shown in Figure 9.
Figure 9: A block diagram of the peripheral hardware controller.
Both the peripheral hardware controller and the CCD control electronics are controlled by an RCA 1802 'COSMAC' microprocessor. The COSMAC system, designed in MIT's Laboratory for Scientific Engineering, is an excellent instrument-control computer: it operates with a simple, low-level, FORTH-like language, and is flexible enough to handle a wide variety of applications.
Figure 10: A view of the entire ETC observation floor, facing northeast. Both ETC modules, each consisting of a manifold and eight CCD cameras, are visible. The CCD control electronics and thermoelectric cooler power supplies are in racks to the north of the modules.
Figure 11: A cross-sectional view of an ETC CCD camera. In this figure, a single Marlow thermoelectric cooler cools the CCD (the tinned-copper braid coupling the CCD heat sink to the cooler has been omitted for clarity). The CCD is shown with its frame-store area obscured by an aluminum cover.
As each pair of thermoelectric coolers generates ~70 Watts of heat during operation, a water/antifreeze solution is pumped through a groove in the back plate of each camera to remove the heat (Figure 11). The antifreeze is added to the solution not only to prevent the coolant lines from freezing in the winter, but also to prevent corrosion in the camera back plates. The solution is circulated through an automobile radiator and air-cooled. The power to the thermoelectric coolers is interlocked to the coolant flow by a McDonnell flow switch, so that the power is turned off if the coolant pump fails. (Without the coolant solution, the camera bodies cannot dissipate the heat from the thermoelectric coolers very efficiently: as a result, camera bodies can heat up to the point where their contents are in danger of being damaged.) With this arrangement, the CCDs typically reach a temperature below -50 degrees C. The moderate dark current in the CCDs at -50 degrees C is not a problem for the ETC, because the typical exposure time is under ten seconds. Also, because the timescale on which the CCD temperature varies is much longer than the exposure time, the CCD temperature is not actively controlled: the CCDs are allowed to get as cold as possible.
The output signal from each CCD is fed over a short cable to a dedicated preamplifier (discussed in detail in Doty, Luppino, and Ricker 1987) mounted on the vacuum manifold. The signal is then passed to the CCD control electronics, where it is amplified and digitized to 12 bits on an analog processing board (one per CCD). In the ETC, the CCD image data is quantized in steps comparable to the readout noise, which is 15-25 e-. The digitized data is then passed to the appropriate TP over a custom high-speed parallel data link.
Figure 12: a) A side view of the ETC manifold (the nearest four cameras have been omitted for clarity). Each CCD camera has rotational freedom about the axis of the camera mounting flange, thereby allowing the declination of each camera to be adjusted individually: once vacuum has been applied, however, the rotational freedom disappears and the declination of the camera is fixed.
b) A view of the ETC manifold from the north.
Each vacuum nipple has one rotateable flange, so each camera can be fixed at any angle in the plane perpendicular to the manifold. Because the manifold is parallel to the polar axis, this rotational freedom means each camera can be set at virtually any declination; however, the declination of the camera must be fixed before the manifold is brought under vacuum. Because of the large field-of-view of the ETC cameras, having all of the eight cameras on one manifold point at the same right ascension can introduce significant overlap in the cameras' field-of-view (just as lines of longitude converge near the poles). The manifold is rough-pumped through one of two additional 2.75" vacuum ports mounted perpendicular to the plane of the camera mounting ports. Two 30 l/s ion pumps mounted directly to 4.5" vacuum flanges on the manifold maintain vacuum in the manifold and cameras. The OC can read the pressure in the manifold over an RS-232 serial line to the ion pump controller. The typical operating pressure in the manifold is ~3x10-6 Torr.
The manifold is mounted directly to a polar sidereal tracking drive manufactured by DFM Engineering, Inc., with the long axis of the manifold coaxial with the polar axis of the drive. The sidereal drive consists of a 35 cm rolled steel disk resting on two roller bearings. The disk can be driven by either a small 1 RPM AC motor (to drive the manifold at the sidereal rate), or a 1 A stepper motor (to slew the manifold). The stepper motor can slew the manifold at a top speed of ~30/minute; the precision of the stepper motor is ~1'. The angle of the disk about the polar axis (and thus the hour angle of the manifold) is measured with a synchro shaft-angle encoder mounted directly to the polar axis of the drive. A synchro-to-digital converter in the peripheral hardware controller digitizes the output of the synchro to 14 bits: the hour angle can therefore be measured to a precision of ~1'. A limit switch on the tracking drive prevents the manifold from tracking indefinitely in the case of a computer failure.
Figure 13: A sample of the record of weather conditions at Kitt Peak reported by the automatic ETC weather station. The record a two-week period in February and March, 1991. The hatching in the plot of the outside temperature indicates times when precipitation was detected. The sharp peaks in the diurnal variation of the outside temperature can be attributed to the fact that the housing in which the temperature sensor in enclosed is in direct sunlight for 2-3 hours each morning.
The design of the roll-off roof was conceived by Stanly Black, AIA, Architect, in consultation with the ETC team at MIT. Several of the more critical design features were supplied by AutoScope, Inc., of Mesa, Arizona, and are based on the successful implementation of an automated astronomical site at the Automatic Photometric Telescope on Mt. Hopkins (reference). A custom roof-control circuit was designed and installed by Pete Manly of Tempe, Arizona: this roof-control circuit allows the roof to be controlled by the OC, through the peripheral hardware controller. A view of the ETC site on the summit ridge of Kitt Peak is shown in Figure 14.
Figure 14: A view from the southeast of the ETC building with the roof fully retracted.
The roof is a simple gabled roof, measuring 3.9 m x 4.8 m and pitched 1:6. It is supported at four points by Osborn V-groove hardened-steel wheels; these wheels run along inverted-V rails on the north and south sides of the building. The V-groove wheels were recommended by AutoScope, Inc., because of their ability to break through any ice that may form on the rails. The roof and wheel well were designed in a way that captures the wheel in the well, thus preventing the roof from lifting off from the building in high winds.
The roof is driven by a 1 HP reversible AC motor mounted to a 50:1 gearbox. A 3/4" pitch steel sprocket mounted to the drive shaft of the gear box drives a stainless-steel chain, which runs over an idler sprocket located ~8 meters west of the ETC building. The chain is connected to the center of the western wall of the roof. Power to the motor passes through a set of solid-state zero-crossing power relays: these relays are arranged in a way that allows the direction of rotation of the motor to be reversed easily. The relays are connected directly to the roof control circuit.
The roof control system was built to act as a computer interface to the roof; however, the safety of any occupants of the building was stressed during its design. The roof control system is equipped with an audible alarm which alerts anyone in the instrumentation area that roof motion is imminent: the alarm sounds for five seconds before roof motion is initiated. There are also two emergency stop buttons in the instrumentation area which, when struck, freeze the motion of the roof and disconnect the roof from computer control.
The safety of the instrumentation was also considered during the design of the roof control system. The roof control system has a built-in watchdog timer which requires a three-second voltage pulse from the peripheral hardware controller at regular intervals: if a pulse does not arrive in time, the control system assumes the computer has crashed, and closes the roof on its own. If the AC power goes out for more than thirty seconds, the control system will close the roof. Finally, the rain sensor is interlocked to the roof motion: if precipitation is detected while the roof is open, the peripheral hardware controller with command the control system to close the roof.
The roof control system initiates roof motion through the relay circuit described above; it senses the roof's location through four limit switches mounted to the rails. The innermost limit switches are intended only to notify the computer that the roof is fully open or closed; the outermost switches are hard limits, beyond which the roof control system will not allow the roof to travel. (Hard stops are welded to the rails in the off chance that the roof control system fails or the roof is moved manually past the hard limits).
The roof control system's computer interface allows the peripheral hardware controller access to all salient information about the state of the roof and the control system. A single computer card in the peripheral hardware controller uses this capability to control the roof motion. This board is the interface to the roof control system, and is therefore responsible for generating open and close commands, as well as the watchdog timer pulses. The roof control board also provides extra safety at the hardware level, so that, e.g., the software cannot ask the roof to open in a rainstorm or drive the roof off of the rails.
6. ETC Software
The ETC is a completely automated instrument, capable of performing an observing program without a human observer on site. The role of the human observer is played by the set of small computers and weather detectors described in Section 5. The Overseer Computer acts as the brain of the ETC, while several peripheral computers control the actions of the ETC instrumentation. Software running in the Overseer Computer controls the flow of ETC operations, from deciding whether observing conditions are adequate to begin work, to whether a candidate transient event was caused by a cosmic ray or some celestial source.
This section is devoted to a description of the "thought processes" of the ETC: the software procedures and algorithms it utilizes in order to perform remote, automatic operations. The operation of the ETC will be described from its highest level, the "day-to-day" decisions the ETC must make about observing times and conditions, to the low-level details of the algorithms used to detect optical transients.
The Overseer Computer (OC) is, as the name denotes, the primary control computer of the ETC (the "brains" of the ETC). As described in Section 5.1, the OC is linked via serial lines to all peripheral instrument control computers, and through these it has complete control of the actions of all of the ETC instrumentation. Software running in the OC which determines the flow of ETC operations and enables the ETC to operate as an automatic, autonomous instrument.
The Trigger Processors (TPs) are responsible for the bulk of the real-time analysis of ETC image data in the search for optical transients. Each TP is a single-board microcomputer dedicated to the analysis of the image data from one CCD camera. The TPs are programmed to perform a select few tasks quickly and efficiently: as such, they act as idiots savants in the ETC system. Each TP runs a basic, low-level operating system stored in an EPROM on board: the software used during operations is downloaded into TP memory before operations begin.
Figure 15: A flow chart of the decisions made by the Overseer Computer at the beginning of a night (see detail in Section 3.1).
During the day, the ETC is idle, and the OC waits for astronomical twilight. At astronomical twilight, the OC checks the weather: if the weather is bad (precipitation has been detected or the measured windspeed is too high), the OC pauses for a short period (~15 minutes) and checks again. If the weather conditions are acceptable, the OC moves the cameras to their starting hour angles (see Section 6.2.4.2), opens the roof, and checks the night sky for clouds (see Section 6.2.4.3 for details). If the sky is cloudy, the roof is closed, the cameras are brought to their idle positions, and the OC pauses, as above; if the skies are clear, observations begin.
Figure 16: A flow chart of an ETC observations cycle (see detail in Section 3.2). The phase labelled "conduct observations" generally lasts ~30 minutes, and is terminated in order to check the observing conditions, store event data, refresh the observing parameters, or, if appropriate, to move the cameras to an new hour angle.
The flow of analysis in the TPs and OC is as follows. Shortly before the integration timer expires, the OC notifies all TPs that new image data are imminent. When the integration timer expires, the OC commands the CCDs to be read out. The TPs then read the image data directly into their onboard memory and begin the process of data analysis, even as the data are being read in. Each TP reports the CCD location of any candidate events it finds to the OC. The OC quickly calculates the celestial coordinates of the event, and compares them to the coordinates of other events detected in other CCD cameras during the same exposure. If the coordinates of two candidate events detected in two cameras coincide to a preset precision (usually ~1 pixel), the event is considered a celestial optical flash.
After the analysis of image data is complete in all TPs, the OC uses the time before the integration timer expires again to notify the appropriate TPs if any of the reported events was real, and that they should store data from such events and any previous events in on-board RAM. The data collected consist of 9x9 pixel subarrays centered on the location of the flash event, as well as 9x9 pixel subarrays centered the locations of eight known SAO stars in the frame, for photometric and astrometric calibration. Because the image frame exposed before the flash event is in TP memory at the time of the detection, image subarrays from the flash location before the event can be collected. Image subarray data is taken from both detecting cameras for the ten exposures following the event detection.
Once the OC has commanded the storage of any event data, it notifies all TPs that new data are imminent, and waits for the integration timer to expire: when the timer expires, the procedure above repeats.
In the sifting process, the pixels in the CCD image are examined individually. The sifting process is divided into five Sift Levels, each of which examines a pixel for a certain signature of an optical transient. All pixels are tested by Level 1 of the sifter: a pixel passing the Level 1 sift test is passed to Level 2. A pixel is tested by successive levels of the sifter until it fails to meet the criteria of one of the levels, at which point it is removed from further consideration and the next pixel is examined. A pixel passing all sifting tests is a candidate optical flash. This procedure is block-diagrammed in Figure 18.
Figure 18: A flow chart of the sifting algorithms in the ETC Trigger Processors (see detail in Section 3.3).
The sifting algorithms used by the ETC are simple and somewhat conservative, in order to eliminate the possibility of not recognizing a real celestial optical flash. Sift Level 1 eliminates all pixels which are not significantly brighter than the sky. Sift Level 2 eliminates all pixels passing Level 1 which have not brightened significantly since the last exposure. Sift Level 3 passes only those pixels which can be considered the center pixel of an optical flash. Sift Level 4 rejects events which occurred in the neighborhood of extremely bright stars. Finally, Sift Level 5 rejects events which occurred in "bad" columns of the CCD (i.e. blocked or bright columns).
The sifting algorithms described here are quite fast: a benchmark test in the laboratory using real CCD image data showed that a full image, even one with an optical transient present, can be fully sifted in under 0.6 seconds. However, this is not the limiting factor in determining the speed of the overall system. When a TP detects a candidate transient event, it reports the location of the event to the OC over a serial line, and then waits for an acknowledgement of reception from the OC before continuing. If there happens to be a large number of candidate events in a given integration, a TP may wait up to 100 ms before continuing its analysis. As a result, the TPs are presently given 3.0 seconds to complete their analyses.
Sift Level 1 checks each pixel to see if it is significantly brighter than the sky. The statistical criteria - - the level of significance required to pass Sift Level 1, N1, and the total noise per pixel, sT - - are supplied to each TP during the calibration phase before observations. The system noise is measured in real time, while the level of significance is set in the OC software at compile time.
In order to increase the speed of Sift Level 1, a comparison, or reference frame is created in TP memory during the calibration phase before observations. The reference frame is an image of the night sky with a constant offset (= N1sT) added: the offset represents how much brighter than the sky a pixel must be in order to be "statistically significantly" brighter than the sky, and thus to pass Sift Level 1.
The image of the night sky is created by median-filtering a CCD image along both rows and columns: first, the median of each column is calculated, stored in a vector, and subtracted from each pixel value; then, the median of each resulting row is calculated and stored in a second vector, which is orthogonal to the first. When the two vectors are summed, the resulting two-dimensional array is essentially the original image minus high-spatial-frequency features, such as stars, and is a good representation of the sky level in the image. Thus, the reference frame is calculated as
where R(x,y) represents the value of the reference frame at CCD coordinates x and y, and, thus, the Sift Level 1 test reduces to
where In(x,y) represents the value of the present image frame (the nth in a series) at (x,y).
The pixels passing Sift Level 1, then, are pixels which are significantly brighter that the sky, and include pixels from most stars in the field and any optical flashes in the field.
To pass Sift Level 2, a pixel must pass the following test:
where In-1(x,y) represents the value of the previous image at (x,y) and g is the system gain, in units of electrons per ADU (analog-to-digital unit). The factor of two in the square root reflects the fact that two independent measurements of total noise are required to make this comparison (to be complete, the term under the square root should read {2sT2 + In-1(x,y)/g + In(x,y)/g}, but as In(x,y) > In-1(x,y) in cases of transients, this can be replaced with {2(sT2 + In-1(x,y)/g)} with no loss of sensitivity).
To minimize the time required to execute this comparison, a table of values corresponding to the right side of equation 9 is calculated during the calibration phase; the table is just
so the Sift Level 2 test reduces to
Thus, the pixels passing Sift Level 2 have brightened significantly since the last exposure.
By passing only those pixels that have brightened more than their eight immediate neighbors (as described above), Sift Level 3 avoids multiple counting of flash events. Sift Level 3 is relatively complicated, and thus relatively time-consuming. However, very few pixels ever reach Sift Level 3, and these pixels are strong candidates for being an optical flash.
As these three criteria described above and the physical restriction that the cameras cannot be pointed more than four hours from the meridian cannot always be met simultaneously, we have developed a pointing strategy which attempts to maximize the scientific return of the ETC. In summary, this strategy is as follows: 1) if the Moon is within 1.5 hours of the meridian, suspend observations, as it is not possible to keep all CCD cameras more than three hours from the Moon; 2) if the Moon is in the western sky, observe in the eastern sky, and vice versa; 3) if the Moon is not up and the Earth's shadow is within three hours of the meridian, point into Earth's shadow; 4) if the Moon is not up and the Earth's shadow is more than three hours from the meridian, point the cameras three hours east of the meridian.
This observing strategy insures that the ETC will conduct observations all night, except for a few nights near full Moon. The restriction of not observing when the Moon is near the meridian has little impact on the quality of ETC observations, as this occurs only near full moon, when the ETC sensitivity is at a minimum. The strategy of pointing into Earth shadow is only effective in the winter months, when the Earth's shadow is at its northernmost point.
The fields-of-view of two cameras, pointed at the extremes of the declination range of the ETC, are examined for the presence of clouds. The cloud-detection algorithm is applied to each of nine 2 x 2 subarrays in each camera in order to detect patchy clouds. If the cloud-detection algorithm indicates the presence of clouds in any of the nine subarrays in either of the two fields, observations are suspended. (The presence of small, patchy clouds generally does not indicate immediate precipitation, but they have an effect on the quality of ETC observations: when they move out from in front of a bright field star, the star will brighten, often by a large enough amount to trigger the ETC event-detection algorithms).
7. Acknowledgements
A large number of people have contributed to the conception, design, development, construction, testing, and operation of the Explosive Transient Camera over the past several years. However, a select few have made significant contributions to the ETC program, both at MIT and at the Kitt Peak National Observatory.
Gerard Luppino designed the thermoelectrically cooled CCD during his work on laboratory x-ray CCD cameras (Luppino, et al. 1987) and the CCD camera for the Rapidly Moving Telescope. During the development of the laboratory x-ray CCD camera systems, Dr. Luppino, in concert with Dr. John Vallerga, adapted existing designs for the CCD electronics to a standard, compact CCD system which is used in the ETC. He also provided innumerable suggestions and comments during discussions about virtually every facet of the ETC CCD camera hardware and electronics.
Initial work on the automation of the ETC, including dome control and weather sensors, was done by Daniel Zachary. The "sifting" software which analyzes CCD images for optical transients was written by Steven Rosenthal. Todd Shauger, Brian Pan, Sheperd Doeleman, and Duane Thresher all contributed significantly to the construction and testing of the CCD electronics; Todd Shauger did much of the work required during the selection and tuning of ETC CCDs. David Breslau was very helpful during the design of the CCD cameras, the manifolds, and the roof motion system. Others who have contributed in some meaningful way are Patrick Mock, George Mitsuoka, Howard Stearns, Leo Rogers, Jim O'Connor, Fred Miller, and Edward Ajhar. We also thank Hans Krimm for a very thorough reading of the manuscript.
We would also like to thank the administration and staff of the Kitt Peak National Observatory, without whose cooperation the ETC would not be possible. In particular, Geoffrey Burbidge, Sidney Wolff, Bob Barnes, Dick Doane, Kurt Cramer, John Africano, John Scott, Hal Halbedel, Bill Binkert, Lee Craven, and Gloria Allen were of great help to our group. Don Versluis was of immense help during the development and installation of the roll-off roof.
Finally, we would like to thank Scott Barthelmy of the RMT team, whose vast store of tools and supplies has saved us time and effort in many situations.
This work is supported by NASA Grant NSG-7339.
Figure 17: A timeline of the flow of operation during a single ETC exposure and subsequent data-reduction cycle in the Trigger Processors. The actual time the Trigger Processor requires to complete the analysis of the data depends on the number of candidate events reported throughout the system, as the Overseer Computer must acknowledge the receipt of each candidate report, and the Trigger Processors wait for the acknowledgement before continuing the sifting process. The tick marks represent one-second intervals.
ETC Paper 3/92
