[time-nuts] Short term 10MHz source

[Magnus Danielson]

Leo,

On 1/9/19 10:10 AM, Leo Bodnar wrote:
> Depends what you call "systematic"...
> 
> I can only speak for Ublox but it is fairly representative of modern (even though its architecture did not change for almost 10 years) GPS chipsets.
> The quantisation runs off internal cheap XO or TCXO that is PLLed to produce MCU core clock that is then phase-accumulation quantised to the current navigation solution target - typically every 1-10Hz.
> Stock Ublox has 26MHz XO/TCXO which is then PLLed by MCU core into system 48MHz clock which runs the Cortex core - including timers - that produce 1PPS TP.
> 
> Somebody measured Ublox 1PPS against the stable timebase and looked at same Ublox reported quantisation correction (ps order of magnitude) and actual error was about 1ns order of magnitude.  This would be unaccounted discrepancy of what Ublox thinks is happening on its 1PPS I/O pin and what is actually there.  This makes analogue delay line correction only viable to that level of accuracy.
> 
> Whether this inaccuracy is the result of a high core PLL phase noise, sloppy quantisation algorithm or hardware I/O drivers is an interesting academic question but won't solve the problem.

Now, what I was talking about was frequency/time-interval counters in
this case, not GPS-chips or GPSDOs. So, your comment seems out of
context in that regard.

However, considering that you have a free-running TCXO, the assignment
of cycle in that clock which is to represent the PPS has a systematic
quantization. The setup is much more complex than a normal counter
setup, so it is otherwise not comparable.

> Ublox core can run off many external clock frequencies from 12 to 40MHz but they all will go through the PLL which makes it pointless to experiment with. 

They better PLL. External clock and steering from the time-product is a
relative straightforward approach if one wants to do that, and then the
PPS quantization will not be as relevant to the steering-loop.

Cheers,
Magnus

> Leo
> 
>> From: Magnus Danielson <magnus at rubidium.dyndns.org>
>>
>> Recall that the quantization is really a form of time-stamp value for
>> the channel in it's relation to the time-base. It's a systematic pattern
>> in the time-base clock and it is phase-locked to the time-base phase.
>> ...
> 
>> So, to conclude, the quantization noise that we have is very systematic
>> in its nature,
> 
> _______________________________________________
> time-nuts mailing list -- time-nuts at lists.febo.com
> To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
> and follow the instructions there.
>