[time-nuts] Clock Project Help

Mike S mikes at flatsurface.com
Wed Oct 28 22:22:46 UTC 2009

At 03:33 PM 10/28/2009, J. Mike Needham wrote...
>Eventually I want to package these for marketing, so they will become 
>specialized once we get a prototype built

You didn't say what level of accuracy you want, but from your 
description ("bedside alarm clock"), 1 second accuracy is way more than 
sufficient (I've not seen an alarm clock with a seconds display, but 
then again, this is the time-nuts list).

If you use Linux, as you mentioned, the kernel RTC should be 
sufficient. You just need to drive a display.

Consider just doing an "ntpdate" as an update mechanism, rather than 
the full ntp steering code. Even with that, you shouldn't be out by 
more than a second even if you only poll every few hours (e.g. 1 second 
drift per 6 hours is a cheap 50 ppm crystal driving the processor).

I'd think that for low-cost volume, you'd just use the chosen processor 
to keep track of time, and output to a display. Most of the time, a 
clock as you described has nothing to do, but count time. So, do that 
by counting cycles in an update loop. Whatever error accumulates when a 
user is making settings will be corrected by NTP at the next update. 
You'd avoid the cost of resources to support even embedded Linux, and 
only have to get network and ntpdate code ported.

More information about the time-nuts mailing list