On Jan 8, 2009, at 2:46 PM, time-nuts-request@febo.com wrote:

Well, like many here, I don't actually have the equipment, especially
the reference std., to do these MDEV, ADEV and other analyses, so,
since I use the GPSDO for a frequency standard and not for UTC, I
thought I'd get the expert opinions. Magnus has several times
indicated here that a TC laying somewhere in and around 100 to 1000
secs is probably optimum. When I enquired some time back about
damping in the TBolt, the consensus seemed to be "leave it at 1.2". I
have, but it just seems to me that won't be optimum for a fixed-
position, lab-located frequency standard -- at the moment, I'm
leaning toward the 0.7to 1.0 area.

Tom's recent chart was quite helpful, especially the 1000 sec curve.
Now, I hope that Tom or someone else follows up on the suggestion to
track performance vs. damping factor. I do understand that the
results for any one GPSDO don't necessarily translate to other
devices, but they don't necessarily don't, either. At least for the
TBolts a lot of us are playing with, one good example (like Tom's)
may well put mine in a better ballpark than the ballpark the factory
wants it to play in, given the factors that you all have described.
Thx everyone for the comments. Look forward to the next round!

Dick Moore

Richard Moore wrote:

Why, since it has been demonstrated that a damping factor of 1.2 is
better than one of 0.7 for a particular Thunderbolt this would tend to
indicate that adjusting the damping without good justification is
somewhat foolhardy.
If in fact the phase noise characteristics of your OCXO are similar toi
the one in the Thunderbolt that Tom measured this would degrade the
performance.

With no way of measuring the effect of such adjustments you are just
hoping that your particular Thunderbolt is similar to the one Tom measured.
Thats not engineering its more like witchcraft.

The probability that you will improve the performance significantly
without a means of measuring the resultant performance is fairly low.
You will never know if either an improvement or a degradation in
performance has occurred.
The one saving grace being that the factory defaults can always be restored.

Bruce

Maybe this should be the subject of a separate thread but to enable
ordinary time-nuts to be able to test their ocxo's and gpsdo's for
phase stability at "home", what would it take as a minimum to be able
to perform something like an ADEV test? This would enable us (the
other half) to see the results of our experiments and tuning of the
gear we have otherwise it is a lot like working blind. I appreciate
that what is normally used is a counter which can continually
timestamp a dut as opposed to a gated counter but what would be the
cheapest way we could achieve this sort of setup?

Thanks and 73, Steve

2009/1/9 Bruce Griffiths bruce.griffiths@xtra.co.nz:

--
Steve Rooke - ZL3TUV & G8KVD & JAKDTTNW
Omnium finis imminet

Dick,

I think you have misinterpreted my postings. I never claimed it was
optimum, or at least never intended to. I think 100 secs is good for
doing additional experiments with damping parameters. It would be
interesting to see just how low the bulb may go. This only since it is
obvious that it makes such a clear difference at 100 secs. It's a choice
out of measurement and interpretation practicality, not optimum from a
use perspective. If you consider my postings you would see that I rather
promote the concept of adjusting the time constant dynamically to
situations rather than say 1234.5678 seconds is the optimum.

Cheers,
Magnus

Steve

If we take TvB's measurements on a Thunderbolt as some guide as to what
to expect:
http://www.leapsecond.com/pages/tbolt-tc/

Then to make meaningful measurements on a Thunderbolt for example one needs:

1. An independent frequency standard with an MDEV better than 1E-12 or
   so for  1 s <Tau<1000 s

2. A means of measuring MDEV with a resolution and internal noise <<
   1E-12 1s < Tau < 1000 s

If one relaxes the Tau range to say 100s < tau < 1000s, then a wider
range of techniques that have adequate resolution are available.
For most GPSDOs the relevant loop time constant will be somewhere within
the (100 - 1000) s range.

One point often missed when quoting/plotting MDEV, ADEV measures is the
measurement system noise bandwidth.
The ADEV and MDEV measures are, in general, dependent on the measurement
system noise bandwidth.
Different systems with different noise bandwidths measuring the relative
ADEV or MDEV of the same pair of OCXOs will produce different results
for ADEV, MDEV.

Possible measurement systems:

1. Phase comparator directly comparing phases of the 2 (10MHz?) sources.
   The system can have a well defined noise bandwidth together with
   adequate resolution if the phase comparator output drives an ADC with a
   resolution of 12 bits or more ( a sigma delta ADC is perhaps the most
   suitable). However the frequencies of the 2 sources must match closely
   and in the case of digital phase detectors the non linearity at the ends
   of the range should be avoided.

2. Heterodyne system where a low noise offset oscillator is used to mix
   down to a beat frequency in the audio range.
   The beat frequency output is low pass filtered and amplified before
   driving either:

A) a sound card  the samples from which are processed to  derive  the
phase  of the beat frequency.

B) A well designed cascaded amplifier limiter low pass filter system
that progressively amplifies the beat frequency signal. The output stage
is a linear comparator and line driver which drives a conventional time
interval counter with a resolution of 100ns or better. Using the beat
frequency output to drive the counter directly results in excessive noise.

3. Dual mixer system with an offset oscillator the performance
   requirements of which are relaxed somewhat because only the differential
   phase shift between the 2 beat frequency outputs is of interest.

Whilst in principle a high resolution (100ps or better) counter with
interpolator could be employed to measure the phase of the divided down
output of the UUT with respect to the standard, the system noise
bandwidth is large and ill defined unless one resorts to crystal and/or
passive RC or LC filters etc with their attendant phase stability problems.

Lacking a suitable frequency standard the best you can do is log the
phase and frequency errors of the thunderbolt when the OCXO is free
running and plot the resultant MDEV.
The best value for the loop time constant should be somewhere in the
close to the value of Tau corresponding to the location of the minimum
value of MDEV.
Perhaps TvB can help by making measurements of the free running MDEV of
a Thunderbolt as measured by the Thunderbolt itself to check the
viability of this method of setting the loop TC.

NOTES:

1. Assembling a high resolution timestamping counter with 100ps or so
   resolution should be reasonably practical.

2. Designing a optimised bandpass slope amplifier limiter cascade is
   relatively straightforward.

3. Optical or equivalent isolation is critical. Where mixers are used
   selecting one which allows the IF ports to be isolated at low
   frequencies is best - Minicircuits have several through-hole models that
   allow this.

4. The real stumbling block is obtaining a suitable reference.
   An FTS1200 or an OSA8607 may be suitable, however these are either rare
   or expensive.
   Some rubidium standards are also suitable.
   TvB only appears to have ADEV plots for the LPRO, however since MDEV is
   somewhat lower than ADEV an LPRO may well be suitable.

5. Using a sound card to timestamp beat frequency zero crossings or an
   equivalent technique is the most flexible and reliable provided that a
   high resolution sound card is used.
   Such a sound card can also be used for phase noise measurements for
   offset frequencies in the 20Hz to 20kHz range.
   Some care is required to keep mains related spurs sufficiently low. I
   have obtained mains related spur levels below 1uV rms by suitably
   arranging the 6m input cables for a balanced input PCI sound card. Since
   this sound card has a full scale input of 4Vrms the effect of 1uV spurs
   is negligible (< 5 fs with 10MHz mixer inputs) for these purposes.

6. A relatively low noise offset source can be assembled from a DDS
   based system provided that a truncation spur free output frequency is
   chosen.

Bruce

Steve Rooke wrote:

Addendum:

Timestamping using a conventioanl gated counter is easily accomplished
using Greenhall's picket fence technique:
http://horology.jpl.nasa.gov/papers/picket_uffc.pdf

The Acam TDC ICs  (http://www.acam.de) have a resolution of a few tens
of ps and a range of up to 200ms or so depending on the chip.
These can easily be interfaced to most micros.

Bruce Griffiths wrote:

Bruce,

Thanks for the detailed rundown. Looking at the picket-fence method,
this looks possible for me but I will have to get hold of the
reference standard. I have a Racal-Dana 1992 with IEEE488 but need to
get an interface card for the PC end. These are fairly cheap to buy.

You spoke about some types of rubidium standards being suitable, would
you care to elaborate on that please? Would something like an Efratom
FRS be suitable?  Generating the picket-fence itself should not be
hard as long as care is taken not to introduce noise. Do you have any
links to articles on the design for the
mixer/zero-crossing/square-wave beat circuit? One question, assuming
that I have a 10MHz reference standard and I'm measuring a 10MHz dut,
how do I arrange for them to be about 1Hz apart, given that we are
measuring for accuracy here? 1HZ different would make the accuracy
1E-7 out anyway, or am I missing something here?

So the real thing for the budget-conscious time-nut seems to be the
reference standard.  The ocxos you spoke about do seem to be on the
rare/expensive side and are an order of magnitude or two better than
the Option 4E I have in the 1992.

73, Steve

2009/1/9 Bruce Griffiths bruce.griffiths@xtra.co.nz:

--
Steve Rooke - ZL3TUV & G8KVD & JAKDTTNW
Omnium finis imminet

Bruce,

Note that my Tbolt time constant plots were made using just a
58503B GPSDO as the reference; not something more exotic.

/tvb

Bruce Griffiths skrev:

I think it is of interest to measure the 10s - 10 ks range.

There is another mistake being done regularly is not to compensate for
frequency drift. ADEV, MDEV and TDEV measures is insensitive to phase
offset and frequency offset, where as drift goes straigh thru, which is
even very easy to show by manually insertion of a linear model into the
respective formulas. On several occasions I have shown that the linear
drift apparent in peoples measurements is where their naive ADEV
calculations "floors out" where as a compensated dataset keeps going
further down. When using a Rubidium as reference for longer taus, such
considerations needs to be done, but it remains purely a post-processing
aspect on the TIE data.

What I have been pondering over is what story does the time deviations
of the ThunderBolt itself says. Just recording that over time and do the
ADEV/MDEV dance would be kind of interesting.

I migth add that it may not be apparent from their datasheets. I have
Minicircuits mixers which is not documented to be isolated, but when
looking at them closely you discover that they are isolated and that
their foot-pattern is also very nice. Some of them is open on the PCB
side so you can actually see exactly how they are wired.

Another thing to recall is that there exist sound cards with more than
two channels, allowing for 3 or 4 channels to be sampled simultaneously,
allowing for a fairly cheap three-corner hat solution among other
things. FFT based cross-correlation techniques can be swooped up with a
few ten lines of wrapping code around FFT libraries such as FFTW for
arbitrary FFT lengths of choice.

It could be added that soundcards having balanced signal input is used
and should be an assumed level.

Another choice for smaller offsets is a standard OCXO. Mixing the output
signal with an external frequency reference (if not being one of the
measured signals) allow for separate locking. Actually, any input could
be used as frequency reference, but if a separate is wanted, it is just
"one more of the same". The locking loop would be the only actual
processing that needs to be done in real time, and it is a fairly
low-intensive processing.

Cheers,
Magnus

I have noticed that Given enough expertise, anything can be made more complicated than need be.

For doing noise testing, there is an option to an expensive osc reference, that has been pointed out many times before. Its advantages is, that unlike other reference standards this one does not have a limit in how low it can measure, and most time-nuts seem to already one or more laying around. The alternative is to just use another one of the same things you are testing (or ANY thing better).
When comparing two independent noise sources, you get an answer that is the RMS sum of the two. That is the answer will be up to 1.414 times the noise of the worse one. It's not too hard to find which is the worse one if you need to with a few more test.

There are also some simple analog alternatives for measuring Phase noise that do not need high resolution Digital TIC, time stamp etc. and can give higher resolution results. I use a XOR phase detector, an analog filter and a radio shack multimeter with PC interface capability.

The ADEV, ODEV and MDEV can then be calculated from the text file data using any of the many great downloadable programs that are available .

The 2G test with a strip chart record of the EFC can be used as a simple way to measure the control loop Time constant and see how the control loop responses to an Osc step function error.

Another interesting and useful effect that can be used if one is careful interrupting the results is the fact that common errors will tend to cancel.
If you compare the noise of two different PLL controlled Osc driven by the SAME 1PPS signal, you will see Just the effect of the control loops and Osc and NOT the effect of the 1pps GPS noise itself. Not what you really want to know when matching an OSC's noise to a GPS signal, but it can provide some interesting insights and results about the control loop and Osc.

I do acknowledge that there are limitations in any of the above and many ways that it can be done wrong, But it can provide a Simple usable test, and in some cases near state of the art testing, for the beginning time-nut that has not yet collected all the great test equipment that is so often referred to.

WarrenS

----- Original Message -----
From: "Steve Rooke" sar10538@gmail.com
To: "Discussion of precise time and frequency measurement" time-nuts@febo.com
Sent: Friday, January 09, 2009 3:18 AM
Subject: Re: [time-nuts] ADEV test setup [was GPSDO TC & Damping]

This would enable us (the

Warren

WarrenS wrote:

The critical requirement is that the 2 standards being compared are
statistically independent.
Comparing a pair of Thunderbolts GPSDOs with similar time constants and
damping will give optimistic results for Tau comparable with or greater
than the loop time constant.
Its is even better is to use 3 or more similar standards simultaneously
logging phase differences between the various pairs (0.5*N(N-1) pairs
for N standards).
It is then possible to obtain estimates for ADEV, MDEV etc for each
standard.

Like all digital phase detectors its best to avoid, if possible, the
nonlinearity inherent at the ends of the range.

Bruce

Tom Van Baak wrote:

Tom

Doesnt that introduce some correlations for larger tau as both the
Thunderbolt and the 58503B are both locked to GPS?

Bruce

I'd like to see a similar test conducted against a local Cs clock (and/or
maser), just to get everything on one graph.

-- john, KE5FX

Steve

The Efratom FRS may be OK, its hard to say without some MDEV measurements.
The specifications only give ADEV for 1s, 10s and 100s.

One way to find out is to compare 3 of them in a 3 cornered hat arrangement.

Bruce

Steve Rooke wrote:

The optimistic results at and above the loop time constant, that results even when 3 or more units are used,
is because the noise is then mostly due to the GPS signal itself and NOT the local oscillators in the GPSDO.
In effect you are then using the same 1PPS signal into each unit, and any common noise on the
GPS 1PPS signal will cancel and not be seen.
So I think what Bruce is saying is that you can not (or is it should not?) use the GPS signal to
measure the GPS's noise.
But the point is, if you want to compare your GPSDO with different settings, or compare it to
another OCXO, It can be done this way, if you do not have a better ref to use.
You could then add the noise of the GPS nose above the control loop time to your
optimistic results if you want true results at high Tau values.

Also note that having the GPS noise cancle is not necessary a bad thing,  It can be a good thing
especially if the GPS noise is not what it is that you want to measure.

Using a phase detector near its end point (or at its crossover point if there is any deadband)
is something that needs to be avoided.
The two basic standard ways to insure that just the center of the phase detector's range is use:

1. Divide the signals down just enough before sending them to the phase detector so that
   the end points is not an issue.  This works when both signals are from devices that are
   locked to a common signal such as the GPS.

2. When one of signals is from a non locked source such as a OCXO whose phase can drift
   any amount overtime, One of ways to limit phase detector issues, and use just the very accurate zero phase point, is to use the Phase detector's output to lock the OCXO in a fast control loop and then by knowing the gain of the EFC input, the filtered EFC voltage can be use as freq drift information to find the ADEV's.

WarrenS

*************:

Bruce

It's on the list.

One earlier idea was to measure tc=1 10 100 1000 10k
simultaneously with 5 Thunderbolts, but I suspected that
unit to unit variations among the GPSDO would partially
cloud the results. So that's why I did back-to-back runs
using the same TBolt, same reference, and same TIC for
each run. All I changed was the TC in the GUI, figuring
that was the safest thing to do.

/tvb

Correct, the choice of reference often has some impact on the plots.

Since I used a GPSDO (a nice one), if the plot extended well
past 10^4 or 10^5 the GPS correlation effects would show up in
the long-term, I think. In that case the plots would start to appear
slightly optimistic.

On the other hand, because I used a GPSDO instead of a maser
reference, the plot I posted appears slightly pessimistic in the
short- and mid-term.

So the main thing about the plot was just how well the difference
between tc=10, 100, 1000 showed up, regardless if the reference
was a maser or a GPSDO. A nice Rb might work too.

That the TC setting was so visible using (only) a GPSDO as a
reference is promising. It means normal time-nuts (oxymoron?)
who don't have cesium or masers lying around have a good
chance to investigate the optimal TC for their GPSDO. Looks
like a 5370 or SR620 would also work for the TI counter.

/tvb

Warren

Another limitation of such phase detectors is that the 2 frequencies
being compared have to be within a small fraction of 1Hz of one another.
This rules out using a low noise reference that happens to have
drifted/aged out of the adjustment range but which is otherwise OK.

Bruce

WarrenS wrote:

Bruce

Tom Van Baak wrote:

Tom

What does the MDEV plot for the particular Thunderbolt look like when
the Thunderbolt itself measures the phase error of the its unlocked OCXO
against GPS?

In other words how reliable a guide is setting the loop TC to coincide
with the value of Tau at the minimum of such an MDEV plot?

Bruce

Steve

Steve Rooke wrote:

The best article I've come across on zero-crossing detector design is:

The Design of Low Jitter Hard Limiters" Oliver Collins, IEEE
transactions on Communications, Vol 44 No 5, May 1996 pp 601-608

Unfortunately its not free, however you may be able to access it via a
Library.

However if you only want to use the technique described in the paper, I
have a couple of spreadsheets that calculate the stage gains and low
pass filter time constants both for the simplified analysis in the paper
and the more general case where the input noise spectral density differs
for each stage.
Some pointers on what to include in the noise calculations for each
stage can be found at:

http://www.ko4bb.com/~bruce/ZeroCrossingDetectors.html
http://www.ko4bb.com/%7Ebruce/ZeroCrossingDetectors.html

Some care is required, in that if the spreadsheet predicts a gain of
less than unity for the input stage, it is in fact better to use a
passive RC low pass filter in front of the first amplifier limiter stage
(without a clamp as typically the IF signal amplitude at the mixer
output is insufficient to cause the clamp diodes to conduct - more
complex clamps are too noisy).
The amplifier limiter chain is then redesigned to accommodate this change.

Don't be taken in by those who would insist that everything should be
linear as long as possible, the resultant deign is suboptimal.
Such comments sprang from the fact that no one at that time had worked
out how to include the effect of the clamps on the performance.
Oliver Collins solved that problem, so there is no longer a valid excuse
for such misguided recommendations.

Bruce

Tom Van Baak skrev:

You could do thia, but in a rotating scheme such that you end up having
measured all thunderbolts for all TCs. That would help decorrelate
individual differences among the thunderbolts.

Cheers,
Magnus

Bruce, thanks, I'm soaking it all up.

73, Steve

2009/1/10 Bruce Griffiths bruce.griffiths@xtra.co.nz:

--
Steve Rooke - ZL3TUV & G8KVD & JAKDTTNW
Omnium finis imminet

WarrenS wrote:

What I am doing to ovoid the "end of range" problem is;
First I divide the signal by two to get a 50% duty cycle. Then when the
phase difference gets to
10% or 90% of the full scale value I switch the phase detector (or
counter) to respond to
to the opposite zero crossing. I keep track of those switches in
software. I use a computer to control things and to keep a log of the
phase difference.

Bill K7NOM

Bill Janssen wrote:

Bill

At this level of precision the waveform duty cycles are never precisely
50% so some correction for this also needs to be made.

Bruce