[time-nuts] Looking for RTC or some combination of MPU/crystal for0.1ppm accuracy.

Bruce Griffiths bruce.griffiths at xtra.co.nz
Sun Sep 13 00:57:50 UTC 2009

However one may be able to set the RTC Interrupt output to 1Hz and use
this (provided that it maintains a fixed time relationship to the
updating of the RTC registers - my need to ask Maxim/Dallas about this).
It would also be possible to probe the RTC registers at various time
delays with respect to the interrupt signal and thereby establish the
time at which the RTC registers are updated sufficiently well to
determine a safe time to read them so that the RTC time returned
corresponds unambiguously to the interrupt transition.
One then uses a continuously running counter to further subdivide the 1
sec time interval between interrupt transitions.
This counter is then sampled at the interrupt transition and by other
events of interest.
These samples can then be processed and combined with the RTC reading
corresponding to interrupt transition that precedes the event the to
form a high resolution time stamp of the event.


Bruce Griffiths wrote:
> The RTC mentioned definitely has a 1 sec time stamp granularity and I
> can see no easy way of improving this as their is no 1Hz output signal
> synchronised to the internal clock.
> Whilst the 32.768kHz output is synchronous with the input to the clock
> divider, this is of little help as there is no easy way to determine the
> phase of the divide by 32768 prescaler used in front of the RTC.
> The 1 sec time stamp granularity probably makes it difficult to
> impossible to correct this system for frequency and phase changes using
> a reference time signal once every 10seconds.
> It would be better to implement a very long (no more than 1 counter
> rollover between successive time reference events, either using hardware
> or a combination of hardware and software) continuously running  counter
> that is sampled by events of interest (including your time reference
> signal).
> You can then process the time reference time stamps and the other time
> stamps to deduce the actual event times.
> One can, in principle, discipline a timer using the reference time
> stamps to produce periodic interrupts synchronised to the reference time
> source.
> You will also need to take into account the various propagation delays
> from the reference time source to each sensor and from each sensor to
> the datalogger system to ensure the various sensors are accurately
> synchronised and, potentially, to avoid data collisions at the data logger.
> Bruce

More information about the time-nuts mailing list