Caution: long OT about power line frequency, why, and how accurate.
If your eyes just glazed over, delete me now!  :)

1) I talked to someone in Engineering at the local utility here in
Vermont, USA.  The line frequency does vary slightly over the course
of the day.  (usually decreasing during times of heavy load).  At
the end of the day, they run slightly faster or slower as needed so
that the total line cycles for the day is EXACTLY 5184000.  That is,
they correct so that the number of line cycles is correct in each
midnight-to-midnight period.

I didn't ask him what they do about daylight savings time :)

2) The basic timebase is atomic clocks of some kind (Rubidium
oscillators, I think).  There is a network (used to be leased
phone lines, maybe satellite, now) of frequency standards which go
from regional master clocks, to local grid master clocks, to
individual generator's controllers.  Most of the slave clocks are
Phase-locked-loop or just amplified copies of the master.  Seems like
they could derive the master oscillator from the GPS signal if they
wanted to set a standard, but as far as I know they are independant
now.

3) Someone asked:
> Why is this done? I mean why does the power line need to be EXACTLY 60Hz?

Most generators are synchronous machines, when they are connected to
the grid, you put torque in and they put power into the line.  If you
load them down, they are motors and draw power from the line.  When
you start them up, if the rotation speed and phase is not exactly in
synch with the grid, they draw huge fault currents until they
synchronize or blow the overload breakers open.  So they HAVE to be
able to synchronize.

Why 60 Hz?  I suspect that Charles Steinmetz et al. at Westinghouse
picked 60Hz as their standard, based on a tradeoff of transformer
losses versus capacitive / resistive losses for their proposed grids
based on the technology at the time.  Maybe not optimal any more,
but there you are.  Why 50Hz in Europe?

Why references?  If each generator synchronized only to the local
grid, there would be phase shifts across the grid and power would
flow haphazardly around the grid, causing losses; and there would be
no control of who delivered power to whom.  The generators control
their speed and phase very carefully to put the desired amount of
power into the grid.  Hence the need for an independent reference
oscillator.  So, as a side effect of needing accurate
synchronization, we have a frequency reference that is widely
distributed, and accurate.  UNfortunately, the resolution at 60 Hz is
poor for many timing tasks.  Oh well.

So: to come back on topic.  For time-of-day clocks, where mains power
is available, you _can_ depend on that for long-term accuracy.
However, you will run slow when the grid is heavily loaded, and run
fast to catch up later.  And be dead when the mains fails.

Our data loggers use watch crystals and calibrate them by tweaking
the load capacitance when needed to get to +/- 1ppm.  ST
Microelectronics makes a real-time-clock chip with cool digital
frequency correction.  (The chip will skip or double pulses to
correct for crystal initial tolerance if you calibrate it.)
If temperature is not varying wildly, they work very well.

If temperature is unpredictable, the crystal temperature
coefficient is a problem.  Another reference is needed for
very accurate clocks when you cannot predict or control temperature.
As mentioned before, time reference transmitters is one option here
in North America, I don't know if there are similar stations is
Europe or Australia, etc.  Another is to put the crystal in a
temperature controlled "oven", if you have the power.  All
interesting engineering tradeoffs.

------------
Barry King
Engineering Manager
NRG Systems "Measuring the Wind's Energy"
barry@nrgsystems.com
Phone: 802-482-2255
FAX:   802-482-2272