[time-nuts] GPSDO using 100Hz
bruce.griffiths at xtra.co.nz
Wed Nov 26 20:33:18 UTC 2008
> I am having trouble keeping my responses from starting new threads,
> If anyone could shed some light on what I'm doing wrong it would be a
> great help to me. It will be interesting to see what thread this post ends up in.
> (Now back to the original subject; I've cleaned up things and added my responses in the text below)
> Re: [time-nuts] GPSDO using 100Hz
> from Tom Van Baak
> Mon, 24 Nov 2008 17:17:34 -0800
> by WarrenS
>>> Besides asking if anyone is using the 100Hz output,
>>> I would like to know why it is that the generally available
>>> GPSDO don't use the 100Hz. I have found that this can
>>> give about 1 ns of certainly with a simple PLL and analog
>>> RC filter, whereas the 1 PPS has more like 100 ns of uncorrected
>>> uncertainty in it, and for the most part the 1PPS needs a processor
>>> to use it instead of simple Phase Lock Loops that I am able to
>>> be used on the 100 PPS
> by /tvb
>> Where do you get your "1 ns" and "100 ns" figures?
>> Have you actually measured these values? Also,
>> over what time average (tau) are you assuming
>> this level of resolution?
> by WarrenS
> The 1ns and 100ns are measured peak values of the
> nominal phase Jitter over a few seconds time as
> displayed on a scope and averaged from a time interval meter.
> The averaged time has to be long enough to let the GPS Phase
> noise cycle thru a few of the Oncore's sawtooth cycles,
> and short enough so as not to include the added received GSP
> signal errors.
> The GPS signal errors that I see are around 10 ns over a 10 plus
> seconds period, with short excursions of an additional 30 ns
> every few minutes generally lasting no longer more than a few seconds.
> These Phase noise values I see are pretty much independent of time,
> as long as the viewed average time is not too long or too short.
> There is also another nominal 50 ns phase error that comes along maybe
> every hour on the 1 PPS and once or twice a day on the 100 PPS both
> lasting from 1 to 10 minutes or so. This happens when the various
> different frequencies involved sync up for a while.
> by WarrenS
>>> On the LeapSecond.com site
>>> It would seem that much of the noise what they have
>>> plotted is the results of NOT setting the GPSDO tracking
>>> TC slow enough. The GPS is adding noise, to the low noise
>>> This is why the Allan Deviation increases by more than 3 db in
>>> the tracking mode at mid averaging times below a 1
>>> 1000 seconds, (15 minutes)
>>> For a low noise Rubidium Oscillator like the SRS PRS10
>>> Bruce early said, "where the ADEV for various GPSDO remains
>>> below 1E-11 over the Tau range of [0.1s, 100,000s].
>>> I said: That Tau has nothing to do with the tracking time
>>> constant that the GPSDO has been is set to, which is usually
>>> recommended to be set for a TC of several hours for optimal results.
> ( for low noise Rubidium Oscillators)
> from /tvb
>> For ideal results start with the TC = the tau where the ADEV
>> of the GPS engine & phase detector crosses the ADEV of the
>> OCXO. However there are other practical considerations.
>> At short averaging intervals when there is a phase difference
>> it is easy for the GPSDO to assume the OCXO is stable and
>> the GPS engine has noise. So you average more samples.
>> At long averaging intervals when there is a phase difference
>> is it easy for the GPSDO to assume the GPS engine is more
>> correct and the OCXO has drifted. So you steer using EFC.
>> At the tau of the ideal TC, the GPSDO sees an average phase
>> difference but can't totally blame either the OCXO or GPS for
>> the error. By definition at this tau, half the noise is due to each
>> subsystem. Hence you nearly always see an ADEV hump that
>> goes above where you'd really like it. Best case sqrt(2).
>> The ADEV hump itself is not indication of a mistuned TC; it is
>> an indication that the GPSDO is working correctly.
>> You can move the hump left and right and distort its shape by
>> changing the TC.
>> In practice I suspect most commercial GPSDO have a TC that
>> appears too low for your liking. I can discuss why if you wish.
> from WarrenS
> Thanks, Very Good information to know and remember, and It
> gives me a much better understanding of what is happing.
> I see that in many of the graphs shown on the Leap Second plots,
> the tracking hump is more like 4 to one instead of 1.5 to one.
> Is this because the tracking TC is too fast for best noise?
> I would love to hear more about it, Please 'Discuss why'.
> Also if you would, I'd like to have a better understand of
> what seems like an over obsessions with Low Noise GPSDO.
> I do understand the need (or at least the desire)
> to have low noise oscillators when using them directly for high
> frequency and/or short time scale data taking applications,
> and the need for a good stable oscillator where it must maintain
> frequency if the GPS signal is lost,
> BUT it sounds like this is not what a large percentage of the
> second hand GPS trackers discussed here are used for.
> It does not seem like it would matter what the noise of the internal
> OSC is, if the unit is only being used for averaging things over longer
> periods such as looking at phase drift over time when doing freq checks.
> Any data taken at a period of say 1000 second to several days is
> pretty much limited by the noise of the GPS and not the Oscillator,
> so why is there the desire for the low noise oscillators in these cases?
Often the use of a low noise OCXO is mandated by unstated applications
1) The OCXO output is used to drive the external frequency standard
input of a counter.
2) The OCXO output is used calibrate other oscillators to 1E-9 or
perhaps a little better and averaging over 1000 seconds or so during
adjustments isnt a sensible option.
3) The OCXO output may be used to control the frequency of a microwave
transmitter and/or the LO of a receiver which requires low drift and
More information about the time-nuts