New Observing Modes for the Next Century
ASP Conference Series, Vol. 87, 1996
T. A. Boroson, J. K. Davies, and E. I. Robson, eds.

Gregory W. Henry

Center for Automated Space Science and Center of Excellence in Information Systems
Tennessee State University, 330 10th. Avenue North
Nashville, TN 37203

Abstract

The Automatic Telescope Instruction Set is a language used to program many current automatic photoelectric telescopes. Intrinsic to this language is a simple dispatch scheduler used to control the order in which observations are made. This scheduler is shown to be adequate in limited situations where a telescope is dedicated to a single long-term monitoring program, but it is lacking for more complex scheduling situations.

Introduction

Robotic telescopes have been used successfully at remote sites for over a decade. A symposium on the capabilities of these telescopes was held in conjunction with the 106th annual meeting of the Astronomical Society of the Pacific at Flagstaff, Arizona in June 1994. Some of the papers published in the proceedings of that meeting (Henry & Eaton 1995) provide examples of the kinds of observations now being made with automated telescopes. For example, Percy and Au, Dukes et al., and Hall outlined a variety of unsolved problems in variable star astronomy that are being addressed by current automatic photoelectric telescopes (APTs). Other papers detailed developments in hardware and software that are enabling continued improvements to these systems. Henry showed how APTs have become more efficient, productive, and economical and how developments in precision photometers and rigorous quality control monitoring have enabled APTs to exceed the precision level of even the best manual observers. Treffers et al. and Markworth described how automated imaging with CCD cameras has become a reality, and Leiker et al. outlined the development of automatic telescopes for education. Other authors discussed plans for networks of automated telescopes, automated telescopes at extremely remote sites such as the south pole and the moon, and the expansion of automated observations to astronomical spectroscopy.

Yet, in addition to the need for successful commercial production of robotic telescopes, one remaining obstacle still hinders the efficient use of these telescopes by large numbers of astronomers. This is the scheduling problem. This paper will show, for the simplest case where an individual telescope is dedicated to a single, long-term monitoring project, that workable, although non-optimal, solutions have been found. However, when many kinds of targets are being observed, each with different and often conflicting constraints, and when multiple observers are sharing a single telescope, automatic telescopes have fallen far short of their potential. Since large, ground-based optical telescopes must be used efficiently because of their high cost and high demand for observing time, and since these telescopes will make greater use of queue scheduling and automated operations, effective solutions to scheduling problems must be found. These solutions will have important benefits to modern observational astronomy.

ATIS Dispatch Scheduling

Tennessee State University (TSU) operates several APTs, each dedicated to a separate long-term project and housed in the Fairborn Observatory site on Mt. Hopkins. In this paper, I describe the operation of one of these telescopes, the Vanderbilt/TSU 0.4-m APT, to illustrate some of the scheduling difficulties encountered with the current scheduling method. The 0.4-m telescope has been photometrically monitoring brightness changes of approximately 150 chromospherically active stars since 1987. Each program star is observed differentially with respect to suitable comparison stars in an observing sequence termed a group. The group observation is specified, including instructions to select the filters, integration times, diaphragm size, etc., via sequences of commands composed in the Automatic Telescope Instruction Set, or ATIS (Boyd et al. 1993), developed at Fairborn Observatory.

The ATIS programming language incorporates a set of group selection rules that control the execution order of groups within a night. These rules operate on parameters specified by the astronomer and located in the header of each group observation request. These parameters include the Julian Date range over which group observations should be made, the hour angle limits that are suitable for the group, the number of times the group should be executed within a night, the moon status (up, down, or either) required for the group observation, and a priority. When the telescope is ready to execute a group, the ATIS scheduler first checks all group requests and determines which ones are currently enabled, i.e., which groups are within the limits specified by the group header. From among the enabled groups, any one of which could be executed next, the ATIS scheduler must select the one group that will be executed next. In a simple winnowing process, the scheduler first considers all enabled groups that have the highest priority level. Within that subset of enabled groups, it looks for the ones with the highest observation request count. From among those, the scheduler selects the group that is closest to the end of its hour-angle window. In the unlikely event that two or more groups are still tied for next execution, the scheduler simply picks the one that appears first in the ATIS request file. When the first group observation has been completed, the scheduler repeats the same process to dispatch the next group for execution and continues doing so for the rest of the night. This scheduling concept, which depends only on internal group selection rules and the current status of the parameters listed in the header, is often referred to as heuristic dispatch scheduling.

This simple ATIS dispatch scheduling procedure has been used successfully for several years to schedule many of the APTs on Mt. Hopkins and has several advantages. First, it is completely robust, i.e., the group selection rules will always converge to the selection of a unique group for execution unless all observational requests have been satisfied. In that case, the telescope simply pauses until additional groups become enabled or the night ends. Second, the scheduler can recover easily from any interruptions due to clouds or equipment failure. Third, a set of groups covering the entire observable sky all can be submitted at one time, and the ATIS dispatching technique will continually schedule the appropriate groups in season for as long as they remain on the observing menu. Fourth, by suitable use of hour angle limits and group priorities, sets of standard stars and quality control groups can be set up around the sky, and these will be interleaved with program star groups at the appropriate times. Finally, as will be shown below, good schedules do result from ATIS dispatch scheduling when the telescope is properly loaded, that is, when the observing requests contain a suitable number and distribution of groups on the sky.

Sample 0.4-m Schedules

The roughly 150 program star groups on the observing menu of the 0.4-m APT are fairly uniformly distributed around the sky and do represent a reasonable load for this telescope. The goal of the observing program is to get one measurement of each group every night throughout each group's observing season. Figure 1 shows the hour-angle track of the telescope across the sky for a night when the load was very nearly optimal. Program star groups are assigned the lowest priority of all groups on the menu, and so are observed only when no higher priority standard star or quality control groups are enabled. The high-priority groups, however, take only a few percent of the total observing time. Therefore, the telescope spends most of the night observing the program star groups, beginning at the west (positive) hour-angle limit at dusk, moving through the meridian (zero hour angle) near midnight, and reaching the east (negative) hour-angle limit just before dawn. In this optimally-loaded condition, the telescope works its way smoothly (except for standard star observations) from the west to the east limits during the night, all groups get maximal seasonal coverage, and the mean airmass of all observations for the night is very nearly minimized, given the requirement that all enabled groups be observed whenever possible.

Figure 1: Hour-angle track of the Vanderbilt/TSU 0.4-m APT for the night JD 2449889. Except for a few necessary excursions to the east and west limits to observe standard stars (marked by filled circles), the telescope moved smoothly from the west to the east hour-angle limits during the 7.5-hr night. The 67 program star groups observed represented an optimal load for the telescope.

However, even in this simple case where the telescope is dedicated to a single monitoring program, optimal loading conditions are not always possible. The actual distribution of program stars on the sky is never perfectly uniform; clumping tends to occur, especially near the Milky Way where the natural density of stars is higher. Removing old program stars and adding new ones of interest usually exacerbates this situation. As a result, some parts of the sky are usually somewhat overloaded, while others areas may be underloaded. On a night when the telescope is overloaded, the telescope will spend a larger fraction of the night in the west trying to observe all groups before they set and will never make it to the east limit before dawn. As a result, many stars in the east go unobserved, so their observing seasons are shortened. Also, the mean airmass of the stars that are observed is much higher than it would be with optimal loading. Since higher airmass results in greater scintillation noise, the photometry suffers considerably.

Figure 2: Hour-angle track of the Vanderbilt/TSU 0.4-m APT for the night JD 2449705. Except for standard star observations (marked with filled circles), the telescope moved smoothly from west to east observing program stars until it reached the eastern hour-angle limit roughly half way through the night. Severe thrashing behavior occurred after that. Although 100 groups were observed during this 11-hr night, many were repeat observations because the telescope was underloaded. The mean airmass of the observations is also higher than on a night with an optimal load.

Underloading presents a similar problem, as shown in Figure 2. With too few groups to observe, the telescope quickly moves through the meridian and over to the east hour-angle limit. If each group has an observation request count of one, the telescope simply waits for additional groups to rise above the east limit, and then observes them immediately. In the example of Figure 2, each group has a request count of two, so that when all enabled groups have been observed once, the telescope moves back to the west limit to make a second observation of some of the groups observed earlier. As soon as a new group rises, however, its group request count of two forces the telescope back to the east limit to observe the new group. Therefore, the ATIS dispatch scheduler causes the telescope to thrash back and forth between the east and west limits, which again results in higher airmass observations and poorer data quality.

The Need for Better Scheduling

ATIS dispatch scheduling, therefore, results in good schedules only when the telescope is optimally loaded. Unfortunately, it is not possible to specify a static optimal load, i.e., a single distribution of stars across the sky that will result in optimal loads throughout the year. There are two reasons for this: (1) the moon's motion across the sky, and (2) the changing length of the night. The ATIS moon parameter is often used to disable some groups containing stars too faint to measure when the moon is up. Also, to protect the photomultiplier tube on the 0.4-m APT and because sky brightness makes stars difficult to measure near the moon, any group within a 20-degree radius of the moon is not executed by the telescope controller. This is actually an override of the ATIS group selection rules, but one that is necessary to protect the photometer. Therefore, moonlit nights often become underloaded. The length of the night also influences the load because that is what dictates the average angular velocity at which the telescope must move from west to east in order to cover the allowable hour-angle window. On short seven-hour summer nights, the telescope must cover each unit of hour angle more quickly than on a longer thirteen-hour winter night. The net result of these two effects is that the density of groups on the sky needed for optimal loading changes substantially throughout the month and the year, and a scheduler that does not account for this from night to night will seldom exhibit optimal loading. Even with APTs dedicated to long-term projects, it is not practical to balance the load on a nightly basis. When new requests are constantly coming in from multiple users and when each user must be treated fairly with respect to the allocation of telescope time, manually adjusting the load each night to give good schedule performance is simply impossible.

Therefore, it is clear that robotic telescopes, even single-purpose monitoring telescopes, need more sophisticated scheduling routines if they are to be used widely and effectively. Drummond et al. (1995) described a new robotic telescope operations paradigm being developed at the NASA Ames Research Center and Tennessee State University that will make it easier for many astronomers to submit observation requests to and receive data from remotely located automated telescopes. In Edgington et al. (this volume), the same group demonstrates the improvements in telescope scheduling that should be realized when this system becomes operational.

Acknowledgments

The development of new scheduling techniques for automatic telescopes is supported through NASA grant NCC 2-883 to TSU. Automated astronomy at TSU has been supported most recently through NASA grants NAG 8-1014 and NCCW-0085 and NSF grant HRD-9550561.

References

Boyd, L. et al. 1993, IAPPP Comm. No. 52, p. 23

Drummond, M. et al. 1995, in Robotic Telescopes: Current Capabilities, Present Developments, and Future Prospects for Automated Astronomy, eds. G. W. Henry and J. A. Eaton, ASP Conf. Ser. No. 79, p. 101

Henry, G. W. & Eaton, J. A. 1995, eds., Robotic Telescopes: Current Capabilities, Present Developments, and Future Prospects for Automated Astronomy, ASP Conf. Ser. No. 79