[time-nuts] backup to GPS after jan 2010, was: OT - GPS and North

Magnus Danielson magnus at rubidium.dyndns.org
Sun Nov 22 23:57:22 UTC 2009


Hal Murray wrote:
> stanley_reynolds at yahoo.com said:
>> Further if at least one cell site has a accurate clock how far could
>> it be repeated before it lost it's useful accuracy using the CDMA
>> signal as in : 
> 
> I think there are two cases.
> 
> The first is if the GPS receiver on a single site breaks, I think it would be 
> reasonable to have that site lock on to it's neighbors.
> 
> The second case is when GPS dies so that all CMDA sites need to coordinate 
> time without help from GPS.  Chains of PLLs are tricky.  If they didn't need 
> GPS they probably wouldn't have used it to begin with.
> 
> I think you could do it if each site had a rubidium or something that was 
> stable enough.  I don't know what "enough" means.  I think the key idea is 
> something like the time constants for distributing the information have to be 
> much faster than the time constant for tweaking the local clock.  That is all 
> the sites have to agree on what to do and understand where they fit into the 
> plan.
> 
> Early SONET ran into troubles with chains of PLLs.  I forget the details if I ever knew them.
> I think they got the time constants wrong and ended up amplifying noise in a certain band
> rather than filtering it out.

Not quite. What happends is that for each stretch and PLL adds some 
noise. At the same time the PLL also has some resonance at its loop 
bandwidth, resulting in a little gain. Say that you have 2 dB of gain in 
each stage and consider that you have 100 such PLLs in series. They have 
by design the same bandwidth and thus their 2 dB of gain occurs on more 
or less the same frequency and you have a 200 dB of jitter gain. 
Simulations done at Bell Labs (using one filter feedbacked throught a 
long cable) made this point painstakingly obvious.

To combat this, a strategy was created where every once in a while a 
narrower bandwidth was applied, thus allowing to filter out the noise of 
the looser PLLs. Specs is set such that there is a healthy distance 
between narrower and broader PLL bandwidths, such that the broader PLLs 
does not provide jitter peaking too close the the narrower PLLs cut-off, 
such that enought of damping slope of the narrower PLLs is in use.

The simple model provides STRATUM-1 (reference clock), STRATUM-2 
(station clock), STRATUM-3 (equipment clock) and STRATUM-4 (line clock).

With narrower PLLs comes higher requirements for stability, which also 
is important for hold-over.

I haven't seen a good reference to all aspects of the SONET/SDH timing 
story, and they are a bit different between SONET and SDH.

Cheers,
Magnus



More information about the time-nuts mailing list