[time-nuts] Frequency standards for different tau in Allen Dev measurement

Taka Kamiya tkamiya9 at yahoo.com
Sat Feb 22 03:55:19 UTC 2020 

It's not like drinking from a fire hydrant.  It's like drowning in hoover dam, get sucked into an inlet, pulverized by turbine blade, and getting spit out into a stream.  

One question :  

You said this: 
"The resolution of your counter tells you about where your 1/tau curve
will cut tau = 1 s, and it goes from there. There is a slight scaling
factor, but if we assume it is 1 for now, it is pretty simple. Your
5335A has 1 ns single-shot resolution, this gives 1E-9 at 1 s, but 1E-10
at 10 s, 1E-11 at 100 s and 1E-12 at 1000 s. You see very clearly when
the linear slope ends and "lands" in the noise, at which time the noise
becomes dominant and is giving you the interesting reading."
By using the same logic, I can keep going up and up on longer gate time and tau keeps getting better and better.  I know at one point, inflection happens and that indicates noise taking over.  But what kind of noise (phase?) and how does that happen?  Only thing that changes in this equation is the gate time.  Everything is constant.  You mean gate time is no longer accurate enough to support the minute shift in phase?
I'm still confused about the precision (not accuracy) of the time base.  Am I still ultimately constrained by this?  Without DMTD, or some kind of pre-scaling of DUT, if I measure Rb with time base using another Rb, they are both rubber-bands, correct?
I'm infinitely curious by nature.  I need to know everything, even to a minute detail, to be satisfied.  I hope you don't get tired of this.    
--------------------------------------- 
(Mr.) Taka Kamiya
KB4EMF / ex JF2DKG
 

    On Friday, February 21, 2020, 9:26:47 PM EST, Magnus Danielson via time-nuts <time-nuts at lists.febo.com> wrote:  
 
 Hi Taka,

On 2020-02-21 23:26, Taka Kamiya via time-nuts wrote:
> I'm sorry, I messed up.  I jumped on more advance topic than I intended.  I'm sure there were answers in the replies but they must have gone way over my head because some of original questions still remain.  I bulletized (is that a word?) the original question with my NEW understanding.  Would someone please respond for me, point-to-point?
No problem. No worries. I hope you end up reading these and the other
replies again and acquire good knowledge. I know it's like drinking from
a fire-hoze, but you did ask some very relevant and fair questions.
>
> 1)  A frequency counter that measures DUT basically puts out a reading every second during the measurement.  When TimeLab is well into 1000s or so, it is still reading every second; it does not change the gate time to say, 1000s.  I understand now, Adev is about phase, not the frequency.  But assuming DUT is sine wave, if there is enough phase change, frequency do change.  I think of phase change as frequency change that is less than full cycle.  So how does counters that outputs every 1 second end up in tau of 1000s?  It will entirely miss phase change that spans more than 1 cycle.

ADEV is about the frequency stability. ADEV can be calculated using
phase or frequency measures. We tend to prefer using phase measures from
Time-Interval Counters for these things.

OK, so let's say that we want to output a counter which provides output
of frequency estimates but for a time-base which is longer than 1 s,
even if we output results every 1 s?

Classically counters could not do that. You acquired a start-value,
waited the time-base, acquired a stop-value, calculated a result to
display and then arm to get a new start-value for the next result. Such
counters will have a limit that the rate of readings will be limited by
the time-base, so if it is set to 10 s, only every 10 s and output is
produced.

To tackle this, one needs a counter that can interleave frequency
measurements, so that it generates new start-points at the update rate
even if the stop-point has not occurred. So, for a time-base of 10 s and
an update rate of 1 s, then every 1 s a new start-trigger is produced,
and then remembered until a stop trigger can be produced, at which time
the start-trigger 10 s back is used to estimate the frequency. In fact,
for this to work, the stop trigger time-stamp is also the start trigger
time-stamp for a new measurement. You can do this with any time-base
really, and the degree of interleaving only depends on the number of
start-points one can keep in memory.

> 2)  I recall reading on TICC manual, in time interval mode, anything that's reasonably good is good enough, because it has time stamp and the count reading.  Clock is used to chunk the data.  Is this still true?  Through this discussion, I ended up with conclusion that there is no inherent advantage over TI measurement when compared to frequency measurement.  Am I understanding this correctly?

There is benefits in time-measures over frequency measures when one
monitors long-term properties. Also, as one tries to create a
phase-curve from frequency estimates, any rounding off errors show in an
slope, as there is a tiny average frequency offset from round-offs. Only
really good such setups does not have significant slope.

>
> 3)  I understand even the BEST counter is only good for Adev nE-12 measurement. Then, with my collection of counters, HP53132A (which averages tons of short period measurement), 5335A (not enough resolution), HP5370A (interval reading is no better than frequency), TICC by TAPR,   Do I even have a chance of doing any meaningful work?  (say work with GPSDO and Rb which some of it does reach E-13)  Yes, I know now, it is NOT possible to do 1 sec Adev but say over 100 seconds?  Right now, I don't have any standard that has adev that good at 1 sec anyway.

The resolution of your counter tells you about where your 1/tau curve
will cut tau = 1 s, and it goes from there. There is a slight scaling
factor, but if we assume it is 1 for now, it is pretty simple. Your
5335A has 1 ns single-shot resolution, this gives 1E-9 at 1 s, but 1E-10
at 10 s, 1E-11 at 100 s and 1E-12 at 1000 s. You see very clearly when
the linear slope ends and "lands" in the noise, at which time the noise
becomes dominant and is giving you the interesting reading. The 5370A is
20 ps single-shot resolution, giving you a whopping 2E-11 at 1 s, 2E-12
at 10 s, 2E-13 at 100 s and 2E-14 at 1000 s. It's some serious
improvement. You are more likely to be limited by your oscillators as
ref and under test at 1000 s with that one, than the instrument itself.

> 4)  Would one person who has infinite patience and experience guide me through getting one reading done correctly with what I already have?  That may include email and phone call.  (I speak English and Japanese)  I don't want to lower S/N of this mailing list by doing this here.  

I think you have contributed by asking some really good questions.

A setup I use a lot is this:

1) Connect a reference oscillator to produce a 1 Hz or 10 Hz signal and
feed into a counter Channel A/TI-start channel. For PPS signals, I make
sure to trigger a but up on the rising edge not to false-trigger. For
some counters this means turning of automatic trigger and set it to 1 V
manually. It is important that no false triggers occurs.

2) Connect a signal under test to Channel B/TI-stop. Adjust trigger to
through-zero or up on the edge as suitable.

3) TI-mode, continuous trigger

4) Collect data in TimeLab, give TimeLab the frequency of the signal on
B-channel, give it the time-base of the period on the A-channel.

5) Look at data as it comes in. Look at phase view, frequency view,
wrapped phase. Look at the ADEV, how the upper end flaps with data, but
how the same tau becomes more and more stable as it comes in.

Using even old counters this setup have helped a lot for many measures.
It is simple and sturdy for many measures. Remember to save traces, to
annotate it carefully so one can understand afterwards what one did.

Using this setup, I swapped a HP53132A (150 ps) for a HP5335A (1 ns) and
then PM6853A (2 ns) to show that a particular problem did not needed the
best counter in the house to be well characterized.

>
> 5)  One time, it was mentioned many of Adev graphs posted are basically a graph of instruments noise graph.  How do I tell when a given reading/graph is exceeding the limit of a setup?  I did do base line where same signal goes to counter's reference input and signal input.  I always have that on my chart so traces does not go below.  Is that enough?
Almost. It's a good start. The first slope for lower taus is due to the
instrument for sure. If you look carefully you will notice that the
actual performance shifts around, because it is more complex than just
being instrument limit, but it is the right ball-park for that part of
the plots. For the upper end, you can be limited by your device under
test drift. This can be handled by simply letting them be turned on
longer. Sub-sequent measurement will have that rising slope move more
and more towards higher taus and thus becomming less like a limit-issue
for a certain tau.
> I appreciate everyone's input.  I am learning a lot but just not digesting well enough.  I'd like to do DMTD after I understand the basics.

Good spirit. DMTD takes some care, but once you learned it, it can be a
magnificent tool.

Cheers,
Magnus

> --------------------------------------- 
> (Mr.) Taka Kamiya
> KB4EMF / ex JF2DKG
>  
>
>    On Thursday, February 20, 2020, 1:41:06 PM EST, Taka Kamiya via time-nuts <time-nuts at lists.febo.com> wrote:  
>  
>  I have a question concerning frequency standard and their Allen deviation.  (to measure Allen Dev in frequency mode using TimeLab)
>
> It is commonly said that for shorter tau measurement, I'd need OCXO because it's short tau jitter is superior to just about anything else.  Also, it is said that for longer tau measurement, I'd need something like Rb or Cs which has superior stability over longer term.
> Here's the question part.  A frequency counter that measures DUT basically puts out a reading every second during the measurement.  When TimeLab is well into 1000s or so, it is still reading every second; it does not change the gate time to say, 1000s.
> That being the case, why this consensus of what time source to use for what tau?
> I recall reading on TICC, in time interval mode, anything that's reasonably good is good enough.  I'm aware TI mode and Freq mode is entirely different, but it is the same in fact that measurement is made for very short time span AT A TIME.
> I'm still trying to wrap my small head around this.  
>
> --------------------------------------- 
> (Mr.) Taka Kamiya
> KB4EMF / ex JF2DKG
> _______________________________________________
> time-nuts mailing list -- time-nuts at lists.febo.com
> To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
> and follow the instructions there.
>    
> _______________________________________________
> time-nuts mailing list -- time-nuts at lists.febo.com
> To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
> and follow the instructions there.


_______________________________________________
time-nuts mailing list -- time-nuts at lists.febo.com
To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
and follow the instructions there.

More information about the Time-nuts_lists.febo.com mailing list