At 11:22 08/06/99 -0400, you wrote:
>Does anyone already did any research about what is more accurate to get
>1Hz division; the 60Hz power line frequency or a cheap 32768Hz crystal?

Short term, the crystal, long term, the line (at least in the UK).  AIUI,
the power companies can easily 'count' the number of cycles that have been
generated, and compare that to a very accurate clock, so in the long term,
the number of cycles will be exactly 50 per second.  However, there is a
permitted variation (short term) of about 5%(?).

>DS-1644 has almost one minute/month error with an internal crystal.

Most digital watches claim accuracy within +/- 15s/month.  That is about
+/-6ppm.  1 minute per month would be about 23ppm, which is far from
unreasonable for a crystal spec.

Nigel