[time-nuts] GPSDO Question

Magnus Danielson magnus at rubidium.dyndns.org
Mon Sep 3 18:36:30 EDT 2007

From: "Poul-Henning Kamp" <phk at phk.freebsd.dk>
Subject: Re: [time-nuts] GPSDO Question
Date: Mon, 03 Sep 2007 20:06:03 +0000
Message-ID: <11565.1188849963 at critter.freebsd.dk>

> In message <c03.175890b6.340db548 at aol.com>, SAIDJACK at aol.com writes:
> >); SAEximRunCond expanded to false
> >This brings up another question: how good are true PID loops that also make  
> >use of the differential term (e.g. correcting for rate of change of the 
> >phase)?  The literature talks about the differential term being hardly used in the  
> >industry because it can add noise and instability to a system...
> I tried it and found that you could indeed et faster convergence, but
> I did it with a HP5370B as phase comparator to minimize the noise.
> With more realistic phase comparators I don't think it is sensible.
> I personally prefer to start with a very short tempered PLL and then
> increase the timeconstant as things get better.

You can get improved result in the early phase from adding a D-term too, but
once you've acheived lock you can reduce it or totally remove it. Just adds
noise when locked.

> The hard part is knowing when to stop increasing the time constant,
> and I have found that keeping track of phase error zero crossings
> is the best indicator:  If you don't see any (for too long time),
> your PLL has too long timeconstant.

This is where the Kalman filter approach has been proven to be better than PLLs
since it will automatically tune itself up. The good people at NIST have
described it.


More information about the time-nuts mailing list