[ntp:questions] How exact are the PPS pulses of common GPSses?

Terje Mathisen "terje.mathisen at tmsw.no" at ntp.org
Wed Feb 13 16:19:43 UTC 2013

Ralph Aichinger wrote:
> Terje Mathisen <"terje.mathisen at tmsw.no"> wrote:
>> Unless you replace the motherboard clock source, anything below ~us
>> precision is wasted.
> Good to know. So basically the 1 or 2µs jitter I get after everything
> has had time to settle down is the maximum archievable unless you
> use rather exotic hardware.

Yeah, more or less.

OTOH the rule of thumb is that you want your measuring gear to have 10x 
the precision of the finished product, so with modern hw resources (i.e. 
not a Pi!) I'd prefer something sub-us.
>> I have seen reports that the original Motorola Oncore UT+ had something
>> like 30 ns RMS jitter (after applying the sw saw-tooth correction),
>> while the $35 Sure board has been tested by one of the time-nuts people
>> at ~25 ns?
> That is great, so basically jitter is a non-issue compared to standard
> computer hardware, and I am pleasantly surprised that the Sure board
> is that good.

So was I!

I have soldered together 3 of them now, they even work inside a server 
room with just a tiny window slice underneath the loading dock overhang.

(Although that is a very marginal antenna placement, it can take many 
hours after a power reset before it syncs up again.)
> Has anybody checked the PPS output for constant shift against
> UTC? Is it possible that the pulses are off with a constant offset,
> even if they have very little jitter? Or is this unlikely from what
> happens inside the receiver?

Not too unlikely unfortunately.

The actual sw clock used for the navigation solution has to be in the 
~10 ns range, since that corresponds to 3m which is a very common 
position accuracy with OK antenna placement.

Getting that sw clock onto a hw signal is more fiddly, the Oncore has an 
onboard 10 MHz crystal which means that the best it can do in hw is to 
pick the closest 100 ns transition, i.e. +/- 50 ns. To get below this 
you have to read and process the "sawtooth correction" signal embedded 
in the serial output.

BTW, since ntp has the ability to adjust for any fixed offset for a PPS 
signal, you should be able to zero out any fixed offset, as long as you 
have a better calibrated source to measure the offset with. :-)

- <Terje.Mathisen at tmsw.no>
"almost all programming can be viewed as an exercise in caching"

More information about the questions mailing list