Recommendations for ASTRO D

There is great interest in achieving ~100 µs accuracy in time assignment for Astro D, in order to improve the chance of detecting a millisecond pulsar in SN1987A and to get optimum results from pulse timing if one is detected. The following suggestions are based on the assumption that Astro D will be operated in a fashion very similar to Ginga. In particular, we suppose that there will be five contact passes each day, with a clock comparison on each contact pass and no clock comparison during each 17-hour segment of remote orbits.

The requirement that the time assignment be accurate to 100 µs over a one-day interval is equivalent to a clock stability of just over one part in 10⁹. This is three orders of magnitude better than the stability of the Ginga clock, and nearly two orders better than the the stability of the temperature corrected clock rate. The preferred method of achieving this accuracy is to control the clock temperature to within 0.002°C (assuming a temperature coefficient for the clock rate of 5 x 10^-7 °C^-1), but given the very tight weight budget for Astro D this option is not likely to be selected. The other method is to carefully monitor the temperature of the clock, and to correct the clock rate for temperature variations. This correction can be applied either in real time, as a hard wired component of the clock circuit, or as part of the data analysis whenever the need for precise timing arises. The two cases differ only in detail, and will be discussed together.

It will be easy to remedy the major problems that we encountered in trying to temperature-correcting the Ginga clock. In particular, there should be no difficulty in obtaining a crystal oscillator with much better linearity in the temperature dependence, and in locating temperature sensors much closer to the clock circuit than was done for Ginga.

In addition, the temperature dependence of the clock circuit should be studied more carefully before launch than was done for Ginga. This of course must be done if a hard wired temperature correction is included in the clock circuit, but it also would be useful for software correction to the clock rate. In principle, the calibration curve can be deduced from operational data (as it was done above for the Ginga clock), but it would be more straightforward to use a laboratory curve. In particular, it will be much easier to check in the laboratory for fine details in the temperature behavior, such as hysteresis (dependence of the clock period on the rate of change of the clock temperature).

The requirements for monitoring the temperature are not particularly severe. If the temperature coefficient for the fractional change in clock rate is 5 x 10^-7° C^-1 (approximately that of the Ginga clock), and the temperature is measured with an accuracy of 0.1°C, then the rms phase drift in 1 minute will be only 3 µs. If the clock temperature is sampled every minute, then the maximum drift between clock comparisons on consecutive orbits will be 21 µs (a random walk of 50 steps forward or backward from each comparison). The rms deviation during remote orbits will be about 75 µs (550 steps forward or backward). If greater accuracy is needed, then better accuracy and/or higher frequency in monitoring of the temperature monitoring can be employed. Long term drifts in the temperature sensors or clock circuit should not be a problem, since they will appear as a linear drift in the clock phase and can readily be removed locally.

If desired, the clock circuit can be shielded from the effect of the temperature day-night cycle with a minor amount of thermal insulation but it the present time this does not appear to be necessary to achieve 100 µs accuracy in time assignment.