> You could use in parallel with your trimmer (and/or regular cap)
> a NTC (negative temperature coefficient) capacitor

This is what I'm going to try - (and I'll look into your suggestion of
an NTC cap as a further experiment)

According to Transpower, NZ's power generator, measuring a
24hr period from 3:30am to 3:30am should be accurate to one
or two cycles, +/- 1/25th second. This is an average, (which varies
because the nominal 50Hz goes up and down during the day, and
that pattern in turn varies with seasons and special events), so the
longer you sample (a year ? "Oi, where's that clock I ordered ?"),
the closer you'd get to 50Hz. But he reckoned 3:30am is about
when they'd put out the 4,320,000th and final cycle for the day.
You could, I suppose, measure a few days and average that

So, I'll use one PIC to count these cycles, putting out a pulse at
the start and end of counting. The start/end pulses of this 24 hour
period are measured by the target PIC with a fine IRQ, maybe
0.001s. It then works out how best to reproduce a second using
a combination of IRQ periods. Any fractional difference in that
reproduction can be sloughed, hopefully unnoticeably, at the minute
or hour increments, spread throughout the day. At 2 cycles original
error per day -> 91s / year, near enough. Present untrimmed
130ppm crystal error -> 4015s / year (20ppm -> 617s / year)

It's not a huge complication and many clocks can be done at
the same time

Besides this I'll be measuring the crystal frequency over a range,
eg 0C to 30C, to see if any temp co, however simple, will be
worth it. If I assume 15C as the average ambient, a clock based
on the above 24hr method will be more accurate than doing nothing

At the very least it's information gathering and playing constructively

-- 
http://www.piclist.com PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist