[time-nuts] GPS noise reduction

Bruce Griffiths bruce.griffiths at xtra.co.nz
Sat Apr 5 19:16:36 EDT 2008

> Bruce,
> The data I am trying to determine is the GPS short-term phase error
> based on the results from comparing the same receiver to multiple
> higher short-term stability sources. If that could be determined
> then you would have an idea of the short-term noise being added
> by the receiver and could possibly correct for it.
> The two oscillators in the system are both quiet, but have opposite
> age rates allowing me to see a difference, otherwise I wouldn't
> be able to tell a difference in the data sets at all. That's why
> additional testing was done to insure the two weren't injection
> locking.
> Thanks,
> Richard

Your assumption that if the 2 oscillators didnt drift in opposite 
directions you wouldnt be able to see any differences in the datasets is 
Unless off course, the time intervals being measured by the TICs are 
sufficiently long that the short term instabilities of the 2 100MHz TIC 
oscillators mask the TIC quantisation noise.

If your oscillators are sufficiently unstable for a fixed isolation 
between the 2 then injection locking will not occur.
The amount of isolation required to prevent injection locking increases 
dramatically as the frequencies of the the 2 tuned circuits approach one 

Independent timestamps of a PPS edge will differ according to the 
timestamp quantisation.
This is entirely independent of the PPS edge to PPS edge jitter either 
random or determinisitic.

With 10ns TIC quantisation there will be variations of up to 20ns or so 
in 2 independent TIC measurements of the same time interval provided 
each TIC uses synchronisers.
If no synchronisers are used then all bets are off.
After averaging this variation will be reduced, but will still be present.
The observed differences will be a combination of the 2 oscillator 
relative instabilities and the TIC quantisation errors.

Averaging the 2 (or more) sets of phase error measurements will improve 
the determination of the short term GPS PPS error by perhaps 40% for 2 
oscillators by 1/SQRT(N) for N oscillators (if and only if TIC 
quantisation noise dominates over the GPS timing noise) however this 
doesnt generate any useful information for more closely disciplining the 
2 (or more) oscillators.

You can achieve the same improvement by increasing the TIC resolution to 
the point where GPS PPS random noise dominates.
However in this case you also generate data that can be used to 
discipline each oscillator more closely.


More information about the time-nuts mailing list