[time-nuts] Using Allan Plots, was(LPRO-101 with Brooks Shera's GPS locking circuit)

Magnus Danielson cfmd at bredband.net
Mon Dec 18 05:23:05 EST 2006


From: "Poul-Henning Kamp" <phk at phk.freebsd.dk>
Subject: Re: [time-nuts] Using Allan Plots, was(LPRO-101 with Brooks Shera's GPS locking circuit)
Date: Mon, 18 Dec 2006 09:58:47 +0000
Message-ID: <40859.1166435927 at critter.freebsd.dk>

> In message <000601c72289$5f9a7350$03b2fea9 at athlon>, "Ulrich Bangert" writes:
> 
> >Having these specs at hand you don't need to make a single measurement
> >to determine the TIC's Allan plot: It will be a straight line starting
> >at 2E-11 @ 1 s and having a slope of -1.
> 
> Not quite.
> 
> 2E-11 is the spec, the counter is likely to do better.

Also, the random jitter is <= 20 ps(RMS) where as the deteministic jitter is
guaranteed to be at (least) 1 ps from the quantization. Tau does different
things to these numbers and it is the combinational effect which forms the
actual ADEV. Toss in other circuit aspects especially for higher taus.

> >If you are eager for making a
> >measurement of your own: Let the SR620 make time interval measurments on
> >pps signals that are derived from its own timebase.
> 
> Wrong.
> 
> This is will not give a random sampling over the internal noisespectrum,
> and the number is likely to be different from the applicable jitter
> value for the specific counter.
> 
> Instead feed in any stable frequency (an ocxo is best) different
> from the clock on the SR620 to both start and stop inputs and measure
> the time interval between them.
> 
> The jitter on that measurement is the number you really want to
> know.

Asynchronous clocks make strange things to circuits. Cross-talk internal to
the counter will show up as a variation in jitter. The intersting thing is that
it will both decrease and increase the jitter depending on the phase
relashionship. This is really a relative time/phase jitter variation. The
amount of crosstalk will certainly give away how well the designers did in
considering crosstalk. The quickest way to disclose crosstalk as an issue is to
have two stable sources be near beating while logging the timings. One of these
may be the time-base. However, since there might be crosstalk from the time-
base you might want to not rely on that too. Single-channel measures will show
channel-timebase crosstalk while dual-channel measures with two independent
clocks will show channel-channel crosstalk. Nonlinearities in interpolators
will also show up here, those are best disclosed by having one of the channels
starting or stopping on the timebase of the counter and the slow phase steps of
the free-running clock will scan the full interpolators range. Preferably the
"free-running" clock is actually a locked and detuned oscillator.

Cheers,
Magnus



More information about the time-nuts mailing list