Previous Article
Next Article
Table of Contents

What Time is it on the Mountain? (1Sep95)
(from KPNO, NOAO Newsletter No. 43, September 1995)

Every FITS header produced at KPNO has lines like the following:

DATE-OBS = '26/06/95 ' / UT date (dd/mm/yy) of observation
UT = '10:57:13.00' / universal time (start of exposure)

How are these values generated and how accurate are they?

The main timebase for the mountain is a WWV receiver that sends coded
information on coax cables strung to each dome. These coax cables
drive the familiar red digital displays in each dome and also drive a
CAMAC or VME module, which is connected to the telescope control
computer and electronics. Thus, from the point of view of the data
taking systems running on the Sun computers, the telescope control
computers can be asked for the UT time just as they can be asked for
the telescope position. The precision of the UT values reported by the
telescope computers is 1 second, although the accuracy of these values
should be substantially better than that. However, the communication
between the data-taking Suns and the telescope computers takes place
over a serial connection that adds significant latency.

The Sun data acquisition computers also have clocks. Currently, these
clocks are synchronized once a day to a standard Sun but otherwise run
free. The magnitude of these daily corrections has been measured to be
less than 5 seconds. Since, under most circumstances, the duration of
the CCD exposures taken with ICE depends on the accuracy of the Sun
clock, we make sure that these clock "updates" take place during the
daylight hours!

When a CCD exposure is started with ICE, a great deal of activity
takes place including---in order---preparing the FITS headers (including
telescope position), preallocation of a disk file to store the image,
asking the telescope control computer for the UT, "preparing" the CCD
and, finally, opening the shutter. Thus the UT recorded in the FITS
header represents the time that CCD target preparation begins, which
is several seconds prior to the time the shutter opens (the
preparation time differs for each chip configuration---including gain
and formatting). Until March 1995, the UT recorded by ICE represented
the time prior to preallocating the disk file; this could, under
extreme circumstances, precede the time the shutter opened by as much
as 30 seconds (although 5-10 seconds was more typical).

IR array exposures taken with WILDFIRE follow a similar sequence of
events. In this case, the last thing done before beginning a series of
exposures is to ask the Sun for the UT. However, due to the
distributed nature of the WILDFIRE controller, data taking does not
commence until 1-2 seconds after the sequence of exposures is
initiated.

In summary, the UT time as reported in the FITS headers of KPNO CCD
and IR array images precedes the initiation of data taking by
"several" seconds. Fortunately, this time difference should be
repeatable to about 1 second.

We are changing things, however, with a goal of providing in the FITS
header an accurate value for the UT that is as close as technically
possible to the beginning of data acquisition. I speculate that for
CCD images taken with ICE, we can reduce the difference between the
real UT of the beginning of data acquisition and the time listed in
the FITS header to be less than 0.1 second.

During the summer shutdown, all the Suns on the mountain have been
changed to use a clock synchronization system called NTP (Network Time
Protocol). NTP, the brainchild of David Mills of the University of
Delaware, enables the clocks on a Sun to be synchronized to UT with an
accuracy of better than 0.1 second through comparisons with accurate
clocks accessed over the Internet. A first order correction for clock
drift is calculated for each Sun's clock so even if the mountain link
to the Internet should go down for a while, the Suns should still be
able to keep acceptably good time. Monotonically increasing time is
also guaranteed. Thus the Suns' clocks should be an adequate timebase.

Changes to ICE are also contemplated to grab the UT value from the Sun
as opposed to the telescope computer (to bypass the latency of the
serial connection and to take advantage of a higher precision clock),
and to move the step of grabbing the UT value from before to after the
CCD preparation step.

Steve Grandi

Previous Article
Next Article
Table of Contents