Introduction

One of the principal capabilities of the Ginga satellite is precise pulse timing of high-flux, short-period X-ray pulsars such as SMC X-1, Cen X-3, and Her X-1. Using data from the Ginga Large Area Counter (LAC), it is possible to determine the pulse phase for these pulsars with an accuracy of 1 ms in an observing time of less than one hour duration. It is therefore important that the assignment of time to the data be made with a precision of better than 1 ms in order to obtain full use of pulse-timing data. The present study was initiated to find out whether the Ginga clock is sufficiently stable to rely on a convenient method of time assignment which uses an average clock rate between consecutive comparisons of the satellite clock with a clock at Kagoshima Space Center. Being able to verify the actual clock rate is particularly important for the remote (non-contact) orbits, for which there are intervals of 16 hours (usually) and even 40 hours (across Sunday vacations) during which there is no comparison with the KSC clock.

We have discovered that there is a large variation in the rate of the Ginga clock of about one part in 10⁶, and that this variation is due mainly to changes in the satellite temperature. However, the dependence of clock rate on temperature is not even approximately linear in the range of satellite operating temperatures, but changes sign after going through a peak near 18°C. This unexpected behavior complicates any attempt to use satellite temperature data to correct the clock rate between comparisons with the clock at KSC.

We have been partially successful in correcting the clock period using satellite temperature data. We have determined an empirical relationship which gives the clock period as a function of temperature. This relationship can be used to convert a time history of temperature into a time history of the clock period. This derived clock period can then be integrated to give the corrected time of observation during intervals when there are no direct comparisons with the clock at the KSC ground station. A large portion of the variations in the clock can be removed by this method. For instance, in a 12-day test interval, the true clock drifts by ±60 ms from that of a constant clock, but the temperature derived clock follows the true clock within ±10 ms.

Unfortunately, the corrected clock determined from the empirical relationship is not perfect. One complication is a day-night cycle in temperature in the course of each satellite orbit, and possible bias in the estimated temperature due to incomplete sampling of this cycle. In principle it is possible to compensate for this bias, but in practice this detail is usually not necessary. A more serious complication is an apparent dependence of the satellite temperature on the satellite orientation with respect to the Sun. To remove this dependence would require a detailed analysis of temperature stratification within the satellite. We feel that such an analysis would result in only a very small improvement in time assignment, and would not be worth the effort involved in carrying it out.

We believe that we are generally able to determine the clock phase from satellite temperature data with an accuracy of 2-3 ms during the remote passes, which is a significant improvement over deviations as large as 10 ms relative to a constant clock rate between comparisons with the clock at the ground station. We show that there is a probable 0.5 ms variation in the clock phase due to the day-night cycle in the temperature, but this is negligible for most purposes. Finally, we discuss the implications of this study for timekeeping on future satellites, particularly Astro D. We suggest that it is possible to use temperature data data to correct the clock rate to a precision of one part in 10⁹, rather than using the standard practice of controlling the clock temperature with an oven.