Eevblog #235 - Rubidium Frequency Standard

[king.oslo]

I am trying to work out how long before the clock deviates from real time by 1 sec.

Please, will somebody who is knowledgeable check my math? This is my attempt.

Clock starts at 10MHz. The drift is 2E-9. So the drift coeffecient is 1.000000002. So I make an intergral to work out how long the clock needs to accumulate 10MHz of drift.

What do you think? Thanks!!!

[jahonen]

I am trying to work out how long before the clock deviates from real time by 1 sec.

Please, will somebody who is knowledgeable check my math? This is my attempt.

Clock starts at 10MHz. The drift is 2E-9. So the coeffecient is 1.000000002.

Thanks!!!

I think the deviation is simply x*(x/(365*24*60*60))*2e-9 = 1, where x is elapsed time in seconds and thus solving for x we get 1.2557e8 seconds which is equivalent to about 4 years. That is assuming that the drift is constant for all t.

Regards,
Janne

[king.oslo]

I am trying to work out how long before the clock deviates from real time by 1 sec.

Please, will somebody who is knowledgeable check my math? This is my attempt.

Clock starts at 10MHz. The drift is 2E-9. So the coeffecient is 1.000000002.

Thanks!!!

I think the deviation is simply x*2e-9 = 1, where x is elapsed time in seconds and thus solving for x we get 0.5e9 seconds which is equivalent to about 15.8 years.

Regards,
Janne

What makes you think so?

Datasheet says 2E-9 drift/year. That is from nominal value? Nominal value is 10MHz.M

[jahonen]

Sorry, corrected a bit previous post on second thought.

Regards,
Janne

[king.oslo]

Janne:

DRIFT             2 x 10-9/year

What do you think this means? I understand you think this is a measure of time, not frequency. Is that correct?M

[jahonen]

I think it means that the frequency may drift 2e-9 per year relative to initial value. So in one year, one could expect the frequency to be between 10 MHz*(1-2e-9) and 10 MHz*(1+2e-9), if it is initially exactly 10 MHz. So the drift is not necessarily constant or in same direction. That is my understanding.

Regards,
Janne

[alm]

I would not expect drift to be constant over time. Aging is a stochastic process (cf diffusion) that will slow down over time, so drift in the first 24h after manufacturing is probably much larger than 24h drift a few years later. A random OCXO (first data sheet I found) is specified as < 5e-10/24h, <1e-8/month and <7.5e-8/year. The same data sheet also states that aging will decrease substantially after the first six months. The difference with a diffusion process is that there may be some systematic contribution, for example degradation of the lamp in the case of rubidium.

[IanB]

What do you think this means? I understand you think this is a measure of time, not frequency. Is that correct?

For small deviations in the order of 1e-9, frequency and time are effectively interchangeable. So if the frequency was off by 1e-9 then the length of a second would be off by 1e-9 also. Which means it would take 1e9 seconds to accumulate a one second error. Which is 280,000 hours, or 11,500 days, or 32 years.

Ha, I think I need to get one of these to use as a time base for a clock. I can make a clock that never drifts 

[king.oslo]

Ha, I think I need to get one of these to use as a time base for a clock. I can make a clock that never drifts 

Yeah. If the train leave a second early, you know whose evaluation to trust

Janne says 4 yrs. IanB says 32 yrs, I say 31622. What should we do?M

[jahonen]

Tip 2 of this Agilent application note might be relevant:

http://www.home.agilent.com/agilent/editorial.jspx?cc=FI&lc=fin&ckey=505277&nid=-33788.0.00&id=505277

I think that the 4 year figure is pretty much surely pessimistic since the drift will be probably less after 1 year (if rubidium standard is kept continuously on), like alm said.

Regards,
Janne

[king.oslo]

I think we are all wrong.

I think a good definite integral (0, ) is:

 (10*((1+(6.341154090044388078630310716550412175015852885E-17))^x))-10.

This will add the annual drift at each second and add each second of drift to eachother.

Does anyone know of tools capable of calculating this? It seems wolframalpha isn't able to

Thanks.M

[IanB]

I think we are all wrong.

I think a good definite integral (0, ) is:

 (10*((1+(6.341154090044388078630310716550412175015852885E-17))^x))-10.

This will add the annual drift at each second and add each second of drift to eachother.

Does anyone know of tools capable of calculating this? It seems wolframalpha isn't able to.

Thanks.M

I think you are pulling our legs. Maybe you need to add a smiley, or people won't get the joke 

[Rufus]

Ha, I think I need to get one of these to use as a time base for a clock. I can make a clock that never drifts 

Yeah. If the train leave a second early, you know whose evaluation to trust

Janne says 4 yrs. IanB says 32 yrs, I say 31622. What should we do?M

And I say 5.63 years.

Assuming worse case drift in the same direction of 2E-9 per year.

Error at the end of T years is T * 2E-9

Average error over a period of T years is T * 2E-9 / 2

The error for 1 second in T years is 1 / (T * 365 * 24 * 60 * 60)

That equality gives T^2 = 1 / (365 * 24 * 60 * 60 * (2E-9  / 2))

my calculator tells me T is 5.63 years.

[king.oslo]

I did the calculation I suggested at the top of the page. I got to 5.631501903 years. Rufus, my result resonates with yours.

I hope we are being very pessimistic here. 5.63 yrs isn't that great.

Thanks.

Kind regards,
Marius

[IanB]

5.63 yrs isn't that great

Do you have any other clocks that accurate?

Anyway, that is a pessimistic estimate on the assumption the frequency drifts linearly by 2e-9 every year. Most likely the drift is not linear and after the device has "burned in" for a year or so it will be much more stable.

[king.oslo]

5.63 yrs isn't that great

Do you have any other clocks that accurate?

Haha! IanB. You are wonderful, man! I love your comments.

No I don't.

The reason I am disappointment is because I'd like one of those atomic clocks that deviate 1sec in 30 million years.

I will have to make a farm of these units, and monitor their drift relative to eachother and relative to another standard.

Thank you for your time.

Kind regards,
Marius

[amspire]

I have a really big problem since I live at an altitude of 1100 meters. Thanks to relativity, time goes much quicker for me then all you lowlanders - about 1 part in 10^-13 times faster.

So do I set my clock to keep in sync with sea level clocks, which means my clock is actually running slow, or do I set it correctly which means my clock will always be gaining time on the sea level clocks. If I set my local clock deliberately slow, then the error filters through to everything else that is time related, like the measurement of capacitance. I hate buying a 1nF capacitor from Jaycar, only to find that when I measure it at home, it only reads as 0.99999999999987 nF.

Richard

[king.oslo]

I have a really big problem ...

I love it man! M

[BravoV]

I hate buying a 1nF capacitor from Jaycar, only to find that when I measure it at home, it only reads as 0.99999999999987 nF.

Wow.. that Jaycar's cap is really good, what part number is that ?   j/k

Btw Richard, I heard you're planning to move all your measurement stuffs upstairs at 1103 meters ? I guess its time to do some re-adjustment again. 

[amspire]

I gather the time dilution difference between shelves on a workbench is measurable - if you work at the American NIST labs. So your frequency counter positioned on the shelf about your signal generator is running about one part in 4 x 10^-17 faster then the generator.

[king.oslo]

Richard, you must compensate for the gravitational fluctuations caused by passing birds or insects.M

[amspire]

Richard, you must compensate for the gravitational fluctuations of any passing birds or insects.M

Don't tell me. It is a waste of time even trying to run my NE555 oscillator when the birds are flying around.

[gxti]

The original video blog got me thinking about buying one of these magnificent devices, but then I talked myself out of it because I really don't need more useless shit. Now this thread keeps taunting me. Inviting me. Oh and will you look at that, I can do even better -- multiple used Trimble Thunderbolt GPSDO on ebay for 100-150USD. Oh, dear. The lure of the Time Nerd is strong

[amspire]

The original video blog got me thinking about buying one of these magnificent devices, but then I talked myself out of it because I really don't need more useless shit. Now this thread keeps taunting me. Inviting me. Oh and will you look at that, I can do even better -- multiple used Trimble Thunderbolt GPSDO on ebay for 100-150USD. Oh, dear. The lure of the Time Nerd is strong

Every now and then I think about adding a circuit to use GPS to improve my rubidium oscillator from 1 part in 10^-9 to one part in 10^-12, but then I have to stop myself and ask "Why?".

To even see a one cycle drift at 10MHz at that precision will take over a day. It is a million times better accuracy then most crystal oscillators in test equipment can manage anyway. It would probably make me think that all my other equipment is lousy.

Next time you think of it, just seek emergency medical help.

Richard.

[joelby]

I'm using GPS to improve the stability of an OCXO (for distributed network synchronisation) and was hoping to compare that to my rubidium oscillator!

It would be far cheaper and easier to just use rubidium oscillators instead of the GPS/OCXO combo, if it weren't for the high power consumption and the chance that the supply could dry up at any moment