[time-nuts] GPSDO for my rubber duckie

Michael Sokolov msokolov at ivan.Harhan.ORG
Tue Aug 2 02:32:13 UTC 2011

Chris Albertson <albertson.chris at gmail.com> wrote:

> I think you can do just fine without GPS and  GPSDXO oscillators and
> the like.  NTP over the Internet is an order of magnitude better then
> you need.

No, I really don't want anything NTP-based.  Or let's put it another
way: part of my objective is to set up a stratum 1 NTP server that would
serve my UTR timescale to the public Internet.  By definition a stratum
1 NTP server does NOT use another NTP server as its reference.

Look guys, I have thought it through and I really want to do it my way.
Sure, it may be way overkill, but I still want to do it my way.  The
reason I have started this thread on time-nuts (rather than the leapsecs
list for example) is not to argue philosophies, but to solicit advice on
the choice of specific GPS receiver / GPSDO types.  I'm still looking
for that.

Forget about my eccentric choice of output timescale and everything else
in this thread.  The part below is what I am soliciting advice for:

I desire to build a circuit that generates an interrupt every
millisecond, and have every 1000th interrupt correspond to the second
boundary in "official" time scales such as UTC, TAI and GPS (which all
agree on where the second boundaries should be).  Forget for the moment
why I want this circuit, let's say I just do.  THIS is the part I am
seeking advice on, nothing else.

Do I need a plain GPS receiver or a GPSDO?  From what I understand,
non-DO GPS units only put out 1 PPS, no 10 MHz, right?  Basically I
would then have to take the 1 PPS input and somehow fill in the
milliseconds in between.  If I do it myself, it seems to me that I would
be reinventing a GPSDO, but perhaps I'm wrong on that.

I don't need the 1 PPS to be phase-locked to 10 MHz, instead I want the
10 MHz to be frequency-locked to 1 PPS.  I don't care if there is no
guaranteed phase relation between the two signals as long as *on average*
there will be 10 million cycles of 10 MHz between successive 1 PPS
pulses.  Nothing bad will happen if any given interval between 1 PPS
pulses features a few cycles too few or a few cycles too many on the
10 MHz, as long as there is no long-term frequency drift between the two
signals.  In other words, I want to be able to clock my circuit from the
10 MHz, catch the 1 PPS once, and then depend on every subsequent 1 PPS
being within a certain fixed window of where I would expect it to be.

Which GPSDO do you guys think would do this job best?

Also what happens to the GPS receiver's serial port in a GPSDO
configuration?  A standard GPS receiver has a 1 PPS output and a serial
port, right?  The serial port transmits a message every second saying
what actual time the next 1 PPS corresponds to, right?  And one can send
commands to that port to change the receiver's configuration, right?
Such as switching between leap-second-corrected and non-LS-corrected
output, right?  What happens to this serial port in a GPSDO
configuration?  Does it pass through GPSDO firmware?  How much direct
access to the actual GPS receiver is still allowed, or none?

These are the kind of things I'm wondering about; I don't need to be
told to use NTP on a PC instead.


More information about the time-nuts mailing list