[time-nuts] Tbolt issues

Lars Walenius lars.walenius at hotmail.com
Fri Sep 2 17:06:26 EDT 2016

You are absolute right in that it is a that depends sort of things. 

Fast temperature changes might really be something that upsets a GPSDO especially if the time constant is long. 

By taking temperature data and multiply with your oscillators temperature coefficient you can do ADEV for the temperature dependence. Quite interesting especially with an HVAC or the sun shining on the GPSDO! Or you can simply take the maximum temperature shift x temperature coefficient for the same time as your GPSDO  time constant to estimate the frequency error you can have depending on the temperature. Of course that is also a that depend sort of thing but at least gives an indication. Say that you have for example a time constant of 1000secs and a temperature change of 2°C during 1000secs and a temperature coefficient of 5E-11/°C that will give you a change of 1E-10 and the loop will not manage to compensate for it in time.


>From: Bob
>Skickat: den 2 september 2016 22:00

>Since the measurement in the frequency domain is a "peak" measure, you need to convert both to frequency error and to an absolute max. If you *do* care about the one second per day (or 10 days) as some do, that is a different factor than one second out of two minutes. Since the noise is likely not to be white noise, the factor is one of those "that depends" sort of things. It includes messy stuff like the room temperature changes and the control loop's response to them. In some designs the response may be multi level. The temp transient hits the voltage reference, DAC, and crystal in the OCXO at different times...


>> On Sep 2, 2016, at 3:08 PM, Lars Walenius <lars.walenius at hotmail.com> wrote:
> I might be completely wrong with my ”quick rule of thumb” (frequency accuracy: 10x the worst ADEV at all Taus longer than the gate time). but my assumptions are these:
> 1. You have a GPSDO. (A free running oscillator as a rubidium or OCXO will not work if that is what you call a ”normal” frequency standard).
> 2. The GPSDO design is such that the frequency error goes towards zero the longer time you measure the frequency. Otherwise you can say nothing from the ADEV to the absolute frequency error I think.
> 3. ADEV can be seen as the standard deviation of all differences in frequency between two nearby frequency measurements (with no dead times) multiplied with a small factor (sqrt 0.5).
> 4. Absolute frequency errors will at least be in the same range as the differences if the long-term error is near zero.
> 5. It will always be larger frequency errors for shorter times. My assumption is for example that at 100 seconds you have 30ppt frequency error that is 3ns drift in 100seconds, If you plot that as straight line you will also have 30ppt at all shorter ”gate times”. If the line isn´t straight you will always find ”points” that has more than 30ppt frequency errors.
> 6. At least around the Taus similar to the time constant of the GPSDO especially the GPS noise will give peak frequency errors compared to the average that are quite high. A factor of ten may be very conservative but as can be seen in the Tbolt measurements by TvB it is quite close. With OCXO and Rb GPSDO´s that I have tried to optimize I have seen about this factor or slightly less. With the DOT050 VCTCXO the ratio was even higher, up to 14 (ADEV 7E-11 1-100secs and 1E-9 max freq errors at 1 second gate time).
> Lars
>> From: Bob
>> Sent: den 2 september 2016 15:20
>> The GPSDO might have an ADEV of 1 ppt at 1 sec and that rises to 30 ppt at 100 sec. It also might not, but let's use those numbers.
>> ADEV is a standard deviation. You can get an idea of the magnitude of the change reading to reading from it. It does not give you a sign for that change. In the case above, it is a good bet that you see strings of 1 sec data with mostly the same sign.  They have to add up to the 30X larger number when tau gets to 100 sec.
>> That creates a major problem if you just look at the 1 sec data. A "normal" frequency standard does not have a rise or bump like that in the ADEV plot. Thus the quick rule of thumb stuff falls apart. The correct way to do it will always be to work out what the noise process is and calculate based on it. That is not a popular thing to do ....
>> Bob
>> On Sep 2, 2016, at 6:38 AM, Lars Walenius wrote:
>> Hello Bert,
>> For me your findings look very much the same as this:
>> http://www.leapsecond.com/pages/tbolt-8d/ 
>> At least for me I should say the (absolute) frequency accuracy for this Tbolt is not better than +-1E10 with 1 or 10 seconds gate times on a counter. Maybe I am totally wrong as both Tom and Charles says that your Tbolt is bad.
>> As Bob Camp said we need a confidence interval. For me in this case it is peak values if they occur more than a couple of times. Think of  voltmeters. If it is out of spec say for only a couple of hours but it happens several times say during a month or year and the voltmeters is not considered bad I should say change the spec. If I were going to use the Tbolt as a reference in a production environment with a 1 or 10 sec gate time I probably would set the internal spec to 2E-10.
>> If I understand correct the Tracor 527E has a low pass filter with a time constant of 1second so the 1 second frequency average data should be relevant.
>> As I am also interested in frequency accuracy with GPSDO´s used with frequency counters, I have a maybe far to simplistic rule of accuracy: (10x the worst ADEV at all Taus longer than the gate time). So for the Tbolt with time constant of 100 seconds it is a hump of about 1E-11 at 100seconds so at gate times shorter than 100seconds I expect to see excursions up to 1E-10. If you set the time constant to 1000seconds and the OCXO in the Thunderbolt is good to 2 to 3E-12 all the way 1-1000 seconds you probably don´t have excursions higher than 2 to 3E-11 at 1-100 seconds gate times.
>> Lars
>> Från: Bert Kehren via time-nuts
>> Skickat: den 1 september 2016 19:09
>> We have been following the Tbolt power discussions but what I am missing is 
>>   the main problem with Tbolts. All the power work will not improve the 
>> frequency  performance of the unit because the frequency is constantly changed 
>> to correct  time. Tbolt is an excellent time device but not good for 
>> frequency reference  past 1E-10. I noticed it when I bought it and compared it with 
>> my Tracor 527E on  the needle and ever since used an Austron 2110 with a 
>> digital 100 sec. loop for  clean up. My Swiss partner Juerg has relied on an 
>> OSA F3 for Tbolt clean up but  continuous bad results on our work resulted in 
>> a detailed analysis using a  HP53132A counter and M100, FTS44060, two 
>> OSA8600's and one of the best FE405's.  The rsult is that the OSA F3 does not 
>> clean up the Tbolt and we see +-4E-11  changes and old data shows even some +-8 
>> E-11 excursions. With the popularity of  the Tbolt an analog or digital 
>> clean up loop would make sense. We are working on  both, the analog because I 
>> saw similar behavior on the FE5680 and FE5650, we did  a GPSDO but do not plan 
>> on using those Rb's but focus on M100 and FRK. 
>> The collective expertise of time nuts could make a significant  contribution
>> For power in critical applications we use the excellent work from Bern Kaa  
>> a friend for the last twenty years and well known because of his published 
>> work  in the European HAM community
>> Bert Kehren
>> _______________________________________________
> _______________________________________________
> time-nuts mailing list -- time-nuts at 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 at febo.com
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.

More information about the time-nuts mailing list