You know Nick, the loop time constant typically used with the HMC1031 loop
filter is typically 5 milliseconds. I'm sure some bigger R's and C's can
used for a longer time constant, and I'm sure that'll help clean up the
awful 10MHz output of the Venus838LPx-T. But it is hardly what I'd call a
"GPSDO".

Tim N3QE

On Mon, Sep 12, 2016 at 2:41 PM, Nick Sayer via time-nuts <
time-nuts@febo.com> wrote:

Hi

The noise performance of the HMC1031 close in is pretty horrible. It’s actually worse than the GPS signal noise. In a normal GPSDO the idea is to use an oscillator that is cleaner at 0.01 to 10 Hz than the GPS.

Bob

Tim wrote:

One point that may not be obvious to people thinking of designing GPSDOs
is this:  The requirement for a very long time constant control loop
(hundreds to thousands of seconds) has nothing to do with the frequency
of the reference signal from the GPS (that is, it is not due to the
reference being a 1PPS signal).  Rather, it is determined by the tau at
which the stability of the GPS signal becomes better than the stability
of the local oscillator (generally an OCXO).  Unless the designer
chooses a very bad local oscillator, this will be in the region of tau =
100-10kS (maybe only 20-50S if using a TCXO instead of an OCXO).

The stability of any higher-frequency signal from a GPS that does not
have a disciplined, high-stability LO (e.g., 10kHz in the case of the
obsolete Navman Jupiter GPS receivers, or 10MHz in the case of the
Venus) will be no better than the stability of the PPS, so the
high-stability local oscillator you add will still need to be
disciplined with the same slow loop you would use with PPS discipline if
you want the sort of results time-nuts expect from a GPSDO.

Best regards,

Charles

I don’t see where you get that.

To generate the phase noise figures (8,9,11,12) they used a Crystek CVHD-950. I don’t think it necessarily follows that a better quality OCXO can’t do better than that, does it?

Ok. Is it the case that the loop filter bandwidth is related to the time constant? Are there methods for minimizing the loop filter bandwidth that might be useful here? I would hazard a guess that obtaining >100s of TC equivalent in pure hardware would be at best difficult. At the same time, it seems like having a microcontroller act as an averager in software between an ADC and a DAC would be just silly.

Jim Miller's 10 kHz GPSDO that’s been referenced here has either solved this problem, or the 10 kHz output of the Jupiter is substantially better than the Venus’ 10 MHz output, or the design doesn’t give the results time-nuts expect from a GPSDO. Which of those applies?

Nick wrote:

They are two ways of expressing the same thing, with the proviso that TC
applies strictly only to first-order, linear time-invariant systems --
in electronics, circuits with with one real pole.

T = RC, where "T" is the time constant and "R" and "C" are resistor and
capacitor values.

So, a single-pole LP filter consisting of a series 10k resistor and a
shunt 100nF capacitor has T = .001 seconds (=1mS).  Its frequency
response (i.e., the -3dB or "corner" frequency) is given by:

f = 1/(2 x pi x T), or 159.16 Hz.

Checking, we note that the reactance of a capacitor is given by x =
-1/(2 x pi x f x C).  When f = 159.16 Hz, the reactance of a 100nF
capacitor is -10k ohms, which is the value that produces a -3dB response
of 159.16 Hz in the LP filter with a 10k resistor.

I have no idea how the Jupiter generates its 10k output -- it was
speculation on my part that its stability must be no better than the
PPS.  I have not looked at the Miller circuit for a long time (also,
note that Mr. Miller published at least two different GPSDO designs, so
one also has to be careful to know which one is under discussion).  I
recall that I did not have much hope that the better of the two would
give TN-quality results, and was very surprised to find that when Tom
tested one, it performed quite well (although, the loop time constant
was clearly way too low -- as it is in the Tbolt with factory default
settings).  I suppose the answer to the question is to be found in ADEV
plots of the Jupiter's 10kHz output.  Perhaps there are some already
published on the web?

There was some discussion of this on the list back in mid-February 2008.

Best regards,

Charles

Figure 1-3 (next page) shows the typical 1PPS performance of the Jupiter-T GPS receiver. The 10 kHz output is also available from the receiver and is phase coherent with the 1PPS signal. This output is made available for functions such as phase locking of crystal oscillators, frequency synthesisers, and similar applications.

-=Bryan=-

Bryan wrote:

Yes, but....  The devil is in the details.

The fact that one thing is phase-locked to another does not necessarily
mean it puts out a good, clean signal.  At short time scales (tau less
than ~100 seconds), the PPS signal from any GPS receiver is noisy.  At
tau = 1second, it is shockingly noisy (~5e-9), and it decreases by a
nominal factor of 10 per decade as the averaging time (tau) is made
longer.  By tau = 1000 seconds, it is pretty respectable (~2e-12 if the
GPS rx designers did their job well).

So, the trick is to use the noisy source (GPS) to discipline the clean
source (OCXO) very gently and very slowly.  That way, the OCXO remains
in control of the output at short tau (< 100 to 1000 seconds), while it
is kept on-frequency over the long term by the GPS.  This requires a PLL
control loop with a very long time constant (equivalently, a very low
cutoff frequency, in the microHertz to milliHertz region).  It is not
practical to build analog filters with time constants that long, so one
must design a digital filter (far from impossible, but not the sort of
thing most hobbyist GPSDO designers are willing to undertake).

If you use a control loop with a short time constant, then the quiet
OXCO just follows the noisy reference source and doesn't improve anything.

If a GPS rx puts out a phase-locked audio or RF frequency (10kHz or
10MHz in the examples we've been discussing), the question becomes
whether that output has better stability (lower jitter) over short
averaging times than the PPS.  The usual way to do this would be to use
a clean local oscillator disciplined in a very slow loop -- the same
thing discussed above, only at 10kHz rather than 10MHz.  To my
knowledge, the Jupiter receivers don't have internal OCXOs devoted to
this, so if the 10kHz output really does have better stability at short
tau than the PPS, it isn't exactly obvious how the designers did it.
Presumably, they would need a very high-Q resonator of some sort.

Best regards,

Charles