Hi > On Jan 18, 2015, at 4:34 PM, Stéphane Rey <steph.rey@wanadoo.fr> wrote: > > Thanks a lot Bob and Magnus for your very helpful comments. > > The HP5370a was indeed in TI mode. By the way what is the difference with +/-TI, the button just aside... > > But I guess I understand where I've missed something : I've tried to put the Rb on channel A and the DUT on channel B but result was always the same but I do understand now that there is indeed a switch to change from COMmon to SEParate and it was always on COM meaning I believe that channel B wasn't used. This explains a lot of things I did not understand. I'm sorry for these so basic issues that might have been solved if I had read carefully the HP5370a manual first. I’ve set that same switch the wrong way several times in the last year. It’s very frustrating …. > > So possible conclusions until now are that I have actually measured the ADEV floor of the system rather than my DUT... which is already nice. The second conclusion from these oscillations seen with the GPSDO under test is that there is very likely in this GPSDO design a systemic noise added to the 10 MHz output (power supply, PCB coupling, ... I'll make further investigations on it later on). > > I will experiment all the suggestions you made and will come back. For information the 1PPS from the HP58503b has a positive pulse width that is only few us length. That sounds about right. > > Now, when considering that the method is to compare the DUT to an other source, I assume then that the other source shall be at least 1 order of magnitude better than the DUT. Otherwise this will be impossible to distinguish who is the instability contributor between the source and DUT, right ? If you can afford a source (or find a source) that’s 10X better, this would be the ideal solution. > > Then the second question is what kind of very stable source can be used to measure DUT which could be Rb or GPSDO which are already in the range of 10E-10 to 10E-12 < 100s ? Your GPSDO is probably around 2x10^-12 at 1 second. Your Rb probably gets to a floor of ~3x10^-13 at a few thousand seconds. An HP5071A with a high performance tube could be used to check the Rb at longer tau’s. A hydrogen maser could also be used. Buying a fully working version of either one will cost you the price of a nice new car (or a couple of new cars). A (rare) BVA OCXO could be used for shorter tau’s. A more practical (lower cost) approach is to compare multiple copies of the same device. With two Rb’s you can get a good guess at the worst they could each be doing. With three of them, you can make a pretty good guess about which one is best / which one is worst. The same approach also works with GPSDO’s. The catch on GPSDO’s is that the filters in them switch time constants automatically, so comparing them in an identical state can be tough. One way to get around problems with things like GPSDO’s is to have multiple standards and use them for different ranges of Tau. OCXO’s have good ADEV close in. Picking up a few and comparing them is pretty cheap. Rb’s have better ADEV at longer tau (> 300 sec). Comparing three or four is not as cheap as with OCXO’s but it’s not totally crazy. Running several GPSDO’s long enough to settle to their longest time constant is another way to go. Lots of choices. Lots of ways to clutter up the basement with TimeNut stuff… Bob > > > Stephane > > > -----Message d'origine----- > De : time-nuts [mailto:time-nuts-bounces@febo.com] De la part de Magnus Danielson > Envoyé : dimanche 18 janvier 2015 16:47 > À : time-nuts@febo.com > Cc : magnus@rubidium.se > Objet : Re: [time-nuts] question Alan deviation measured with Timelab and counters > > Bonjour Stéphane, > > On 01/18/2015 03:37 PM, Stéphane Rey wrote: >> Hello, >> >> First, please do apologize for the confusion answering in the bad email. That's things I'm absolutely able to do when replying at 3 am ! Again, sorry for that and thanks Magnus for having corrected this. > > Ah well, that's water under the bridge now. I only mentioned it for Bob's reference. > >> Back to my setup : >> >> There is indeed nothing on the STOP input of the HP5370a. The standard 10 MHz comes from the GPSDO HP-58503B and feeds the HP5370a Standard input. Its ADEV is given on page 240 of that document : http://www.leapsecond.com/museum/hp58503a/097-58503-13-iss-1.pdf We see that the shape is starting at about 2E-12 at 1s, increase to 2E-11 at 100s before decreasing again down to E-13 for above 10E3s... >> >> The setup #1 was using the Racal DANA Rb connected on the START input >> which is specified at E-9 / E-10, given on page 16 of the manual : >> http://bee.mif.pg.gda.pl/ciasteczkowypotwor/Racal/9470-9479.pdf >> The EXT input receives the 1PPS from the HP58503b. It apparently drives both the START and the STOP of the acquisition (the two lights are blinking and the time between two measurement is no longer adjustable from the front panel RATE potentiometer and the period between two samples is 1.0s (detected by Timelab). > > If you run the counter in frequency or period mode, you normally use the STOP input, which is then internally split to the START and STOP channels. > > If you run the counter in TI mode, then they are usually separate, but you can force them the same using the START COMMON switch. > > We tend to use the TI mode, with two basic setup: > > Stoopid simple: PPS to START and measured clock to STOP. This setup has the down-side that the jitter of the PPS (which can be much higher than that of the clock) can dominate, if so, the next setup is relevant: > > Standard setup: PPS to ARM/EXT input to trigger measurement. DUT to START channel and reference clock to STOP channel. Sometime the clocks is interchanged, sometimes it is important, somtimes not. > > Record the TI data. > >> But yes, the ADEV plot sounds really strange as it goes incredibly low >> after few seconds which is not consistent with the stability of the >> sources I'm using which is why I felt something was wrong > > OK, you made what we call a instrument noise limit measurement. Then you do the same thing as a normal measure, but you have start and stop channels see the same signal split. It may be good to let the stop channel has a meter or two of additional coax to de-correlate the rising edges. This setup will let you measure the effect of white noise, slew-rate and counter resolution. It can be good for fault analysis and see if the setup gives reasonable noise or if you can improve it. > Adjustment of the trigger points will select a point of optimal slew-rate (and sometimes avoid false-trigger noise) and thus finding the optimum trigger noise. > > Squaring up the signal may be a nice way to improve the setup. > > Anyway, such setup has the 1/tau plot behavior and that was what I saw. > The fact that you kept going down was a clear clue that you where doing such a setup rather than doing a suitable delta. > > Now, try the two setups I proposed, letting the STOP channel being delayed with about 1 meter extra cable, and record the result. Do share for comments. Then, using the setup giving the lowest trace for measure your two other sources as DUT. > >> On Setup #2 I've only replaced the Racal Dana Rb with the GPSDO to >> test. I've not made this design and not checked yet anything on it. >> Could these oscillations be from power supply noise ? To be checked. >> But how can it follow the ADEV plot of the Racal Dana Rb ? mmm.... >> Coincidence is not something I like too much and I believe something >> is clearly wrong in my measurement >> >> But what ??? > > Re-arranging the setup and it will be interesting to see both these setups. Then we can start making some comments on that result. > >> On the Timelab setup screen before launching the acquisition I've left all the parameters as it without touching them. I've just seize 10E6 in the frequency field. > > Usually that's all that is needed. > >> Ah, by thay Magnus, for the downmixed test I've forgotten to change this value, I will check on monday when back at the office. > > If you only have your TIM file with you back home, all you have to do is to press (e) to Edit the trace, as I recall it. I might have edited the file directly also. When doing that, I helped another time-nut at one time. > > Uncheck the "Use Input Frequency" and then input 10 MHz (or whatever) to "DUT Frequency". > > To actually make gains from a mixer-setup, you need to do more processing to filter and square up, but for the moment, it's just a nice lab-exercise. :) > > Cheers, > Magnus > _______________________________________________ > 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. > > > > --- > L'absence de virus dans ce courrier électronique a été vérifiée par le logiciel antivirus Avast. > http://www.avast.com > > _______________________________________________ > 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.

> One way to get around problems with things like GPSDO’s is to have multiple > standards and use them for different ranges of Tau. OCXO’s have good ADEV > close in. Picking up a few and comparing them is pretty cheap. Rb’s have better > ADEV at longer tau (> 300 sec). Comparing three or four is not as cheap as with > OCXO’s but it’s not totally crazy. Running several GPSDO’s long enough to > settle to their longest time constant is another way to go. On that subject, I should probably mention that the current TimeLab beta (at http://www.miles.io/timelab/beta.htm ) has a realtime N-cornered hat display. I needed that feature to measure a bunch of OCXOs against a pair of Corby's trick 5065As, so I finally bit the bullet and added it. It is very cool, subject to the (many!) caveats with separated-variance measurements in general. This feature should work with measurements taken by counters as well, but I haven't tested that yet. Even worse, I won't have time in the immediate future to document it very extensively. Short version: measure three devices against each other, and then use the 'e'dit dialog to assign Source A and Source B labels to each of the resulting three plots. Go to the ADEV or other xDEV measurement view and hit ctrl-h to toggle the N-cornered hat display mode. The program will use your assigned source labels to separate the individual source variances. It won't work if you don't spell the source names consistently between each pair of measurements; if any of the measurements have different t0 intervals, bin counts, or trace history settings; if you don't read the mouseover help text for the Source A/B fields in the trace properties dialog; or if you don't read everything Bill Riley has written about 3-corner hat measurements. TimePod users can assign the source labels at acquisition time, via the new fields in the Advanced tab. Read the new mouseover text for the Stability and Ch0/Ch1/Ch2 fields _carefully_. Users of counters, or those who want to render existing files in 3-cornered hat mode, will need to add source labels in the 'e'dit dialog. Symmetricom/Microsemi 3120A users should email me offline if interested. I don't have a beta for the 3120A app yet, but it will be there eventually... -- john, KE5FX Miles Design LLC

John, Your new three-corner hat feature is really cool. Already tried it, even if I did not spend quality time on setting source labels correctly. Whenever I have a setup capable of running it, I will test-spin it again. I suspect that when you make separate measurements, the noise of the individual sources does not match up perfectly which may reduce quality, especially systematics like hum. Cheers, Magnus On 01/19/2015 03:23 AM, John Miles wrote: >> One way to get around problems with things like GPSDO’s is to have multiple >> standards and use them for different ranges of Tau. OCXO’s have good ADEV >> close in. Picking up a few and comparing them is pretty cheap. Rb’s have better >> ADEV at longer tau (> 300 sec). Comparing three or four is not as cheap as with >> OCXO’s but it’s not totally crazy. Running several GPSDO’s long enough to >> settle to their longest time constant is another way to go. > > On that subject, I should probably mention that the current TimeLab beta (at http://www.miles.io/timelab/beta.htm ) has a realtime N-cornered hat display. I needed that feature to measure a bunch of OCXOs against a pair of Corby's trick 5065As, so I finally bit the bullet and added it. It is very cool, subject to the (many!) caveats with separated-variance measurements in general. > > This feature should work with measurements taken by counters as well, but I haven't tested that yet. Even worse, I won't have time in the immediate future to document it very extensively. Short version: measure three devices against each other, and then use the 'e'dit dialog to assign Source A and Source B labels to each of the resulting three plots. Go to the ADEV or other xDEV measurement view and hit ctrl-h to toggle the N-cornered hat display mode. The program will use your assigned source labels to separate the individual source variances. > > It won't work if you don't spell the source names consistently between each pair of measurements; if any of the measurements have different t0 intervals, bin counts, or trace history settings; if you don't read the mouseover help text for the Source A/B fields in the trace properties dialog; or if you don't read everything Bill Riley has written about 3-corner hat measurements. > > TimePod users can assign the source labels at acquisition time, via the new fields in the Advanced tab. Read the new mouseover text for the Stability and Ch0/Ch1/Ch2 fields _carefully_. Users of counters, or those who want to render existing files in 3-cornered hat mode, will need to add source labels in the 'e'dit dialog. > > Symmetricom/Microsemi 3120A users should email me offline if interested. I don't have a beta for the 3120A app yet, but it will be there eventually... > > -- john, KE5FX > Miles Design LLC > > > > _______________________________________________ > 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. >

> John, > > Your new three-corner hat feature is really cool. Already tried it, even > if I did not spend quality time on setting source labels correctly. > > Whenever I have a setup capable of running it, I will test-spin it again. > > I suspect that when you make separate measurements, the noise of the > individual sources does not match up perfectly which may reduce quality, > especially systematics like hum. True, and Bill Riley's documents emphasize that simultaneous measurements are best. That's easy to do with the SMA jacks on the TimePod but more of a challenge with counters, unless you buy several of them and run them from separate GPIB adapters or serial dongles. IMO, if you can keep correlation, tonal artifacts, and other systematic errors under control and make long-enough measurements to generate a lot of data points in the tau range of interest, then the technique should work well with non-simultaneous acquisitions. At the end of the day it's no better or worse than the quality of the individual ADEV measurements. You have to be wary of measurements at taus where any of the traces have large error bars or are wandering up and down over time, or where some of the traces are suspiciously close to each other. -- john, KE5FX Miles Design LLC

John, On 01/20/2015 01:55 AM, John Miles wrote: >> John, >> >> Your new three-corner hat feature is really cool. Already tried it, even >> if I did not spend quality time on setting source labels correctly. >> >> Whenever I have a setup capable of running it, I will test-spin it again. >> >> I suspect that when you make separate measurements, the noise of the >> individual sources does not match up perfectly which may reduce quality, >> especially systematics like hum. > > True, and Bill Riley's documents emphasize that simultaneous measurements are best. That's easy to do with the SMA jacks on the TimePod but more of a challenge with counters, unless you buy several of them and run them from separate GPIB adapters or serial dongles. > > IMO, if you can keep correlation, tonal artifacts, and other systematic errors under control and make long-enough measurements to generate a lot of data points in the tau range of interest, then the technique should work well with non-simultaneous acquisitions. At the end of the day it's no better or worse than the quality of the individual ADEV measurements. You have to be wary of measurements at taus where any of the traces have large error bars or are wandering up and down over time, or where some of the traces are suspiciously close to each other. Agree. For oscillators, they should have been turned on long enough such that any drift is negligible. Alternatively you process out the quadratic trend out of it. The later should be accompanied by some quality measure of how much remaining systematics there is (see Jim Barnes PTTI paper on Drift Estimators). Also, identifying systematic noises (such as in phase-noise plot) and notching them out before processing should improve the quality of the 3/N-cornered hat processing, since it really is for solving power-sums of random noise, but not for systematic noise. Until such processing of each measurement is in place, we have to make as clean measurement runs as possible, which is the basic quality of the ADEV. Cheers, Magnus

		Attila Kinali

Moin Magnus, On Wed, 21 Jan 2015 07:07:54 +0100 Magnus Danielson <magnus@rubidium.dyndns.org> wrote: > For oscillators, they should have been turned on long enough such that > any drift is negligible. Alternatively you process out the quadratic > trend out of it. The later should be accompanied by some quality measure > of how much remaining systematics there is (see Jim Barnes PTTI paper on > Drift Estimators). Do you mean "The measurement of linear frequency drift in oscillators", http://tf.nist.gov/timefreq/general/tn1337/Tn264.pdf ? Attila Kinali -- It is upon moral qualities that a society is ultimately founded. All the prosperity and technological sophistication in the world is of no use without that foundation. -- Miss Matheson, The Diamond Age, Neil Stephenson

Hej Attila, On 02/06/2015 01:22 PM, Attila Kinali wrote: > Moin Magnus, > > On Wed, 21 Jan 2015 07:07:54 +0100 > Magnus Danielson <magnus@rubidium.dyndns.org> wrote: > >> For oscillators, they should have been turned on long enough such that >> any drift is negligible. Alternatively you process out the quadratic >> trend out of it. The later should be accompanied by some quality measure >> of how much remaining systematics there is (see Jim Barnes PTTI paper on >> Drift Estimators). > > Do you mean "The measurement of linear frequency drift in oscillators", > http://tf.nist.gov/timefreq/general/tn1337/Tn264.pdf ? Yes. I use the PTTI link http://tycho.usno.navy.mil/ptti/1983papers/Vol%2015_29.pdf I was just lazy to dig the link up when I wrote that. Cheers, Magnus