Thanks guys for the input,
I wanted to clarify that the time interval being measured at each
GPS 1PPS is typically 2/5 of a 1.6us window (Rb/16) or 640ns at
setpoint. The filter input value is built by accumulating (not
averaging) 120 of these samples and sawtooth correcting the result
before filtering. This gives a filter input value with 10ns per
sample / 120 samples or 83ps per count resolution per 120-second
update. The original 16-bit filter was extended to 24-bits to
improve the filter resolution with the 23-bit DAC. The disciplined
oscillator stability using this inexpensive design is orders of
magnitude better than the 1e-11 per day specification of the
original Shera design.
I am just powering up a similar version with a variable update
rate on an MTI 260 double oven OCXO to see what the stability looks
like with a respectable OCXO. (I finally found one MTI 260 out of
six I tested that didn’t jump in phase every few weeks!) Once I
have this unit up and stable and have some baseline data I will
pull out my Vectron and Bliley 100 MHz OCXOs and try Ulrich’s
suggestion of using one of them for the TIC clock to see if that
improves the disciplined oscillator stability. They both have
around 1e-9 stability and should be enough better than an XO to
see if further improvement in disciplined oscillator stability
is possible using this design.
The reason I went this route was to see what improvements could
be made to create a high performance $50 controller using readily
available DIP packaged parts that could be assembled on a perf
card over a weekend by the average hobbyist. Age can do bad things
to the eyes and hands that make surface mount components and high
density IC pinouts hard to deal with for some older hobbyists.

Thanks again,
Richard

-------------------------------- Original Message ---------------------------------
Subject: Re: [time-nuts] Metastability in a 100 MHz TIC
From:    "Tom Van Baak" tvb@LeapSecond.com
Date:    Fri, July 20, 2007 6:57 am
To:      "Discussion of precise time and frequency measurement" time-nuts@febo.com

Partly yes, for tau < 1 second.
Mostly no, for tau > 1 second.

Perhaps I misunderstand your setup, but it seems to the
third timebase is only used for a time interval measurement,
not the time measurement. Thus the requirements on its
stability are much less than the two 1pps sources. Think of
it not as a "timebase", but a "time interval base".

For example, suppose you want to measure 1pps sources
which are within 10 microseconds in phase, to a resolution
of 100 ps. To make an ADEV plot for tau 1 second to 1 day,
you need to collect days of data, but you only need a
TIC timebase that is accurate and stable to one part in ten to
the 5th, at tau of 10 us. Any cheap XO will do that, no?

You don't need cesium timebases for a 1pps TIC.

/tvb

time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.

Richard,

A 1.6 us window mean you have almost no issues with
the accuracy or stability of your 100 MHz sample clock.
10 ns out of 1.6 us is 1/2 percent; clock counts won't
exceed 160; a quartz timebase is overkill.

Do I understand correctly: you make each raw 1pps
time interval measurement down to 10 ns resolution,
then (in software, I presume, one second later) apply
a negative sawtooth correction with 1 ns resolution, then
average 120 of those sums, and then expect a 84 ps
resolution result? Something doesn't sound quite right.

/tvb

----- Original Message -----
From: "Richard H McCorkle" mccorkle@ptialaska.net

Thanks guys for the input,
I wanted to clarify that the time interval being measured at each
GPS 1PPS is typically 2/5 of a 1.6us window (Rb/16) or 640ns at
setpoint. The filter input value is built by accumulating (not
averaging) 120 of these samples and sawtooth correcting the result
before filtering. This gives a filter input value with 10ns per
sample / 120 samples or 83ps per count resolution per 120-second
update. The original 16-bit filter was extended to 24-bits to
improve the filter resolution with the 23-bit DAC. The disciplined
oscillator stability using this inexpensive design is orders of
magnitude better than the 1e-11 per day specification of the
original Shera design.
I am just powering up a similar version with a variable update
rate on an MTI 260 double oven OCXO to see what the stability looks
like with a respectable OCXO. (I finally found one MTI 260 out of
six I tested that didn’t jump in phase every few weeks!) Once I
have this unit up and stable and have some baseline data I will
pull out my Vectron and Bliley 100 MHz OCXOs and try Ulrich’s
suggestion of using one of them for the TIC clock to see if that
improves the disciplined oscillator stability. They both have
around 1e-9 stability and should be enough better than an XO to
see if further improvement in disciplined oscillator stability
is possible using this design.
The reason I went this route was to see what improvements could
be made to create a high performance $50 controller using readily
available DIP packaged parts that could be assembled on a perf
card over a weekend by the average hobbyist. Age can do bad things
to the eyes and hands that make surface mount components and high
density IC pinouts hard to deal with for some older hobbyists.

Thanks again,
Richard

-------------------------------- Original Message ---------------------------------
Subject: Re: [time-nuts] Metastability in a 100 MHz TIC
From:    "Tom Van Baak" tvb@LeapSecond.com
Date:    Fri, July 20, 2007 6:57 am
To:      "Discussion of precise time and frequency measurement" time-nuts@febo.com

Partly yes, for tau < 1 second.
Mostly no, for tau > 1 second.

Perhaps I misunderstand your setup, but it seems to the
third timebase is only used for a time interval measurement,
not the time measurement. Thus the requirements on its
stability are much less than the two 1pps sources. Think of
it not as a "timebase", but a "time interval base".

For example, suppose you want to measure 1pps sources
which are within 10 microseconds in phase, to a resolution
of 100 ps. To make an ADEV plot for tau 1 second to 1 day,
you need to collect days of data, but you only need a
TIC timebase that is accurate and stable to one part in ten to
the 5th, at tau of 10 us. Any cheap XO will do that, no?

You don't need cesium timebases for a 1pps TIC.

/tvb

time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.

time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.

Tom,
The GPS 1ns sawtooth corrections are accumulated in a 16F688 during the
120 second sample period with care that they match the same 120 1-second
10ns resolution phase samples collected. The accumulated sawtooth
correction is read at the end of the sample period before the sawtooth
correction for the next sample is sent by the GPS, scaled to match the
accumulated phase count resolution and added to the phase count before
the value is sent to the filter. This simplifies the design and has the
same effect as adding a scaled sawtooth correction matching the counter
resolution from each sample once per second. I do indeed make sure both
numbers cover the same samples and have the same LSB resolution before
adding so the results are valid.

Richard

Tom Van Baak wrote:

Tom

If the quantisation error of the sawtooth correction is random then
averaging will reduce the noise in the average sawtooth correction to
somewhat below 1ns.
If the 100MHz oscillator phase drifts randomly with respect to the phase
of the OCXO being disciplined then averaging will indeed reduce the
noise of the phase measurement to below 10ns, however although the
calculated resolution is 83ps the noise in the result will be somewhat
higher than this.

Indeed a quartz timebase is perhaps too stable for effective averaging
unless the PPS signal has around 10ns or so rms jitter on its leading edge.
However if a less stable timebase is used to ensure it has sufficient
short term instability to ensure accurate averaging it will be necessary
to measure/calibrate its frequency perhaps every second to correct for
drift in the timebase oscillator.

Relying on the 100MHz crystal oscillator phase drifting around in a
random fashion so that the phase error averages are unbiased estimates
of the true phase error is perhaps expecting too much unless heroic
measures are taken to ensure that the 100MHz oscillator doesnt phase
lock to the OCXO output via injection locking.

The phase error averaging will be improved by adding sufficient jitter
to the PPS signal to increase its random timing jitter to around 10ns
rms or so.
If this is done the 100MHz clock can be derived from the oscillator
being disciplined and the averaged phase will still have a very small bias.

Bruce

Bruce,

I like your point about the random quantization error in the
sawtooth. Yes, that would help the noise by a few dB.

On the other hand it would also seem the 10 ns resolution
of the TIC is the limiting factor (by an order of magnitude)
over the 1 ns resolution limit of the sawtooth corrections,
so improving the quality of the sawtooth corrections has
limited gain.

Now, I'd still like to pursue the issue of noise in the 100 MHz
oscillator. Do we agree one doesn't need a cesium for this?
Or even an XO?

True, you want some accuracy in the 100 MHz. But the counts
are only integers from 0 to 160 so the accuracy requirement
is just 3 digits, 0.1% (so cheap quartz, at 1e-6, or cesium, at
1e-13, is extreme overkill). I mean, almost anything wiggling at
100.0 MHz will serve as an adequate timebase.

Also, as you point out, instability or jitter is your friend, not
your enemy in this case. Would it be possible to introduce
the +/- 5 ns jitter deliberately in the 1pps trigger level instead
of in the timebase? I.e., slow down the rising edge enough
so that you get jitter for free?

Another solution might be to deliberately choose an inaccurate
and unstable oscillator; use the 1pps to count oscillator cycles
per second, as well as to count the time interval. The larger
count can be used to calibrate the smaller count on every count.
This gives all the jitter you need and avoids any injection issues.

While you're at it, how about N of these oscillators, each making
its own out of phase measurement of the same OCXO-GPS 1pps.
Another couple of dB of resolution...

/tvb

----- Original Message -----
From: "Dr Bruce Griffiths" bruce.griffiths@xtra.co.nz

Tom
Tom Van Baak wrote:

As long as one keeps track of the frequency drift of the timebase
oscillator this is true.

Yes adding stochastic jitter to the leading edge of the PPS signal is a
good way of injecting the required noise into the measurement.
The most predictable way is to slow down the PPS edge (a simple RC
filter should be more than adequate ) and feed it into one input of a
comparator whilst the other input is connected to a noise source.

Yes a suitably noisy LC oscillator should work, will probably need to
reduce the tank Q somewhat to achieve sufficient noise/jitter.
Deliberately using resistors to introduce predictable noise may be a
useful technique.

Attenuating the sinusiodal ouput of an oscillator that is too stable
using a resistive attenuator with a high output resistance may also be a
viable technique for producing a sufficiently noisy clock.

This is getting way to complicated, surely the method of using the
inherent noise in the hardware corrected PPS signal in conjunction with
a D flip flop acting as a simple precedence detector (as proposed
several weeks ago) is easier, cheaper and simpler, it also has more
resolution than almost any other method one can devise. It can easily be
elaborated slightly to allow detection and rejection of phase error
measurement outliers.

Clock the D flip flop with the hardware corrected PPS signal, connect
the divided down OCXO (or other standard) being disciplined to the D
input. Interrupt the microprocessor on the trailing edge of the PPS
signal, read the Q output of the D flip flop (the 200 milliseconds PPS
width of the PPS signal from an M12M or similar GPS timing  receiver is
more than adequate to allow the D flipflop to settle with an extremely
low probability of being in a metastable state- even a few microseconds
is probably more than sufficient) and add 1 to the phase count whenever
D is 1 subtract 1 whenever D is 0. The EFC voltage is then adjusted to
keep the phase error at zero (corresponds to a 50% probability of the D
flipflop output being 1 when read by the micro). This simplistic
algorithm can be replaced by a more optimum algorithm as required.

This can all be built with a few DIP ICs to ease assembly, however a
well designed 2 layer PCB with a ground plane is probably advisable.
Other than a programmable delay line and a high resolution DAC no
unusual parts are required.

Instead of using one microprocessor to do everything several of simpler
processors each dedicated to a particular function may be better.
One processor could be dedicated to hardware correction of the PPS
signal, whilst another can implement the phase error measurement and the
OCXO control loop.

Bruce

From: "Tom Van Baak" tvb@LeapSecond.com
Subject: Re: [time-nuts] Metastability in a 100 MHz TIC
Date: Fri, 20 Jul 2007 17:10:45 -0700
Message-ID: 00fe01c7cb2b$95ca8be0$0300a8c0@pc52

Certainly. Rather, it is a mostly uncorrelated signal rather than random, but
the effect is the same never the less.

I was thinking the same thing.

The TIC only needs a low tau for the range of time on which we depend on it,
which is up to 1 second, which is the maximum time between two PPS signals.
A TIC which does move around will cause small scale errors, but since we have
fairly good sources we can avoid that problem by having the TIC use one of
those. But the TIC does not need to have very good ADEV at tau = 10 ks or
100 ks, it should just not be so high that the scale error grows to high to
cause a problem. So, in practice I agree with you, but the ADEV values does
have an effect even over there in the ADEV spectra, it is just that they work
differently.

LC Colpit oscillator should be able to do it. :)

We know that we want an increment of 10, so a simple BCD counter setup with a
simple resistor-DAC setup creating a sawtooth which ads ontop of the PPS with
the same amplitude should be able to do the trick. Clocked with the PPS, so
we will not need blitzing speed here. HC should do just fine.

The two-minute averaging should work out very nicely with that. The resolution
would not actually improve, but the hanging bridges would, so you would get the
10/120 ns resolution.

Even very simple interpolating schemes would give much better resolution for
less buck. An improvement of 1 or 2 decades comes very cheaply. Just look at
the HP 5335A for instance. It has 1 ns resolution couting everything except
events in 10 MHz. The interpolators gives a gain of 200.

Cheers,
Magnus

Alan Melia wrote:

Alan

Using a slow unsynchronised sinewave is not the way to go a noise source
is better and is easily implemented.
Essentially the technique used by HP in one of their counters would
suffice, phase modulate say a 10MHz signal by a few degrees and then
multiply the output by 10
to produce a 100MHz signal with 10x the phase modulation. This
simplifies the design and construction of the phase modulator. A diode
double balanced mixer can be used to phase modulate a signal by the
required amount. Feed the LO port of the mixer with the signal to be
modulated apply the modulation signal to the IF port and add the output
of the RF port in phase quadrature with the original signal.

The drawback is the complexity and the fact that the resolution is still
inadequate to achieve the maximum performance from the better GPS timing
receivers with a good antenna and site. The simpler and cheaper D
flipflop precedence detector used together with hardware sawtooth
correction has far higher resolution. It also has the advantage of not
requiring any high frequency clocks.

Bruce