[time-nuts] Project on precise timing over Ethernet

Magnus Danielson magnus at rubidium.dyndns.org
Sat Sep 12 15:08:16 UTC 2009

Hal Murray wrote:
>> Actually, you can see this as a Shannon information channel, analog or
>> digital.
> What sort of bandwidth do I need to run a PLL over a long link?
> I assume it takes enough to cover all the sources of error:
>   drift in the master
>   drift in the local oscillator
>   drift in the communication link

noise in the communication link, includes systematic/deterministic noise 
such as DJ, ISI, head-of-line packet delays etc. etc.

effective sampling rate (some systems, such as SDH/SONET, have 
non-linear buffer mechanisms which makes updates rate and quantization 
size large), assuming a justification mechanism or asynchronous clock 
transfer mechanisms.

effective quantization size, assuming a justification mechanism or 
asynchronous clock transfer mechanisms.

There are many phase modulations sources, some taking the form of 
variations and other takes the form of biases.

> Does it depend on my target accuracy?  Do I need more or less bandwidth for a 
> better answer?  Am I even asking a sensible question?

It is a sensible question, but you need to ask it the right way for the 
particular type of system. Also recall, you can trade bandwidth for 
resolution. You need to understand the systems information capacity 
need, and you need to understand how it deteriorates. For most things, 
you rarely discuss it in those terms, but experience show that early 
naïve reduction in capacity may hurt alot later to overcome.

> What sort of changes in delays to I get on 10 km of fiber or coax or twisted 
> pair?  Is temperature the only interesting variable?

Temperature is not the only interesting variable, but it is very 
important. The group delay over fiber depends on length of fiber, 
temperature and frequency of the ligth (aka "wavelength"). Fiber is by 
nature dispersive and this causes frequency-dependent group-delay. Fiber 
changes physical length with temperature, but also the wave equation 
changes and thus shifts the response. A majority of group delay change 
in the actual fiber comes from this temperature change of the 
dielectrum. The lasers being used also change frequency with 
temperature, and the shift accumulates over the distance.

> Can I measure tides with a 10 km link?  Or is that swamped by temperature?

Not with a normal system, but an inventive mind could possibly do that. 
I think there is more effective systems for that.

> How about microwaves?  Does the speed of light in air change with temperature 
> or humidity or whatever?

It changes. Wind shaking the towers should not be forgotten.

> What's the final bandwidth out of the control loop on something like a GPSDO?

Notice how that bandwidth may not directly correlate with that of a 
transmission channels as I talked about.


More information about the time-nuts mailing list