Sneaky Errors
I have two thunderbolts, set up so that I can switch over to the backup unit if the primary fails.
All is well with that, but what could I do to detect a less obvious failure, like 9.999999 MHz output?
If they disagree, I don't know how to resolve which is correct.
The thunderbolts produce status information and error estimates about how
they are doing. If you are willing to trust that, you can remove the one
that is falling out of tolerance. If you aren't willing to trust that, then
I'm pretty sure you'll need a third frequency standard to compare each to.
Thanks,
-JP
On Wed, Oct 19, 2011 at 3:46 PM, David VanHorn <D.VanHorn@elec-solutions.com
wrote:
I have two thunderbolts, set up so that I can switch over to the backup
unit if the primary fails.
All is well with that, but what could I do to detect a less obvious
failure, like 9.999999 MHz output?
If they disagree, I don't know how to resolve which is correct.
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.
On Wed, Oct 19, 2011 at 12:46 PM, David VanHorn <
D.VanHorn@elec-solutions.com> wrote:
I have two thunderbolts, set up so that I can switch over to the backup
unit if the primary fails.
All is well with that, but what could I do to detect a less obvious
failure, like 9.999999 MHz output?
If they disagree, I don't know how to resolve which is correct.
The only way is to have third or fourth 10Mhz oscillator. If you are very
conservative or if this is for some critical requirement then I'd get
something other than another TB. It is always remotely possable that the
reason for out of spec performance is a design flaw. Having three identical
units would never catch this kind of problem as it would hit all three at
the same time. Using that same line of logic. Maybe your
third redundant standard should not depend on GPS. I'd say you might want
a Rubidium. Then you compare all three periodically check that the two TB
track each other and that the Rb drifts slowly away. Later if ever the two
TBs fail to track each other you will need all that drift data you collected
to figure out which TB is "correct" because it's only 50/50 that the correct
TB would be the one closest to the Rb.
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.
--
Chris Albertson
Redondo Beach, California
This is the famous "man who has two watches does not know what time it is"
problem.
Lucent solved it for telecom with the RFTGm (Reference Frequency and Timing
Generator) equipment, consisting of GPS disciplined OXO and Rubidium
oscillator modules that continuously checked each other via 1 PPS and 10 MHz
links. The frame housing the oscillator units selected one or the other for
distribution to six connectors. Sadly, the output is 15 MHz.
Otherwise, the only solution is more watches, preferably by different
manufacturers.
In some ways, this reminds me of the ancient parity check for memory
locations. Parity is not used anymore for commercial computers, because a
memory error is usually not alone, and errors soon make the computer lose
its way and halt. In this case, plan on the TB either running correctly or
failing due to some alarm. Alarms must be monitored, of course.
You could set up a time interval counter to show the phase between the two
outputs. The Racal-Dana 1992 does that at 10 MHz.
Out of curiosity, what would be the consequences of a steadily increasing
phase error? Would it offend your sense of perfection or would it have real
consequences?
Bill Hawkins
-----Original Message-----
From: David VanHorn
Sent: Wednesday, October 19, 2011 2:47 PM
I have two thunderbolts, set up so that I can switch over to the backup unit
if the primary fails.
All is well with that, but what could I do to detect a less obvious failure,
like 9.999999 MHz output?
If they disagree, I don't know how to resolve which is correct.
Out of curiosity, what would be the consequences of a steadily increasing
phase error? Would it offend your sense of perfection or would it have real
consequences?
Phase error wouldn't bug me. My worst fear is that the 10.000000 MHz standard might be 10.000002 MHz.
I need to implement something that is at least relatively simple or best case, automatic.
On 10/20/2011 12:18 AM, David VanHorn wrote:
Out of curiosity, what would be the consequences of a steadily increasing
phase error? Would it offend your sense of perfection or would it have real
consequences?
Phase error wouldn't bug me. My worst fear is that the 10.000000 MHz standard might be 10.000002 MHz.
I need to implement something that is at least relatively simple or best case, automatic.
Steadily increasing phase error is to let there be a frequency error.
Frequency is the derivate of phase, so it comes with the territory.
So a 200 ns per second phase drift would provide a frequency error of 2
Hz on your 10 MHz. Can't have one without the other.
Cheers,
Magnus
Steadily increasing phase error is to let there be a frequency error.
Frequency is the derivate of phase, so it comes with the territory.
So a 200 ns per second phase drift would provide a frequency error of 2
Hz on your 10 MHz. Can't have one without the other.
I understand, I'm just saying that if the absolute phase is wandering a bit over tens of seconds, it's NOT an issue.
On 10/20/2011 12:57 AM, David VanHorn wrote:
Steadily increasing phase error is to let there be a frequency error.
Frequency is the derivate of phase, so it comes with the territory.
So a 200 ns per second phase drift would provide a frequency error of 2
Hz on your 10 MHz. Can't have one without the other.
I understand, I'm just saying that if the absolute phase is wandering a bit over tens of seconds, it's NOT an issue.
So, you are saying that the stability is not that important. Fair
enough. I also assume that absolute phase biases is not of your
concern... so that your second is shifted by say 4711 us doesn't hurt
you, as long as you get your 10 MHz on average.
What frequency stability do you need, over which time? (i.e. what ADEV
value for which tau)
Cheers,
Magnus
Seems to me that 200 ns is 720 degrees of phase error, which is a lot.
A person handy with logic circuits could build a simple phase detector
with a flip-flop and an RC filter with a tenth second time constant.
An analog circuit could detect 360 degree rollover and set off alarm
bells. Note that you still have the two-watch problem.
Two equal divider chips ahead of the flip-flop could allow larger errors
before rollover. The error may reverse itself and run the phase error down,
and then reverse again as the two ovens cycle at different rates. This
would be normal behavior, unworthy of an alarm.
An additional challenge would be to build logic to select the oscillator
output to be distributed, then compare the output to the output of three
oscillators in three phase detectors. The device that had phase rollover
would put itself in standby, alarm, and leave you with the two watch
problem.
Perfection demands many oscillators with a voting system. Long winter nights
could be spent solving these problems. I'm too old for that stuff.
(I also post the most recent ideas first, so as not to reread old ideas.)
Bill Hawkins
-----Original Message-----
From: David VanHorn
Sent: Wednesday, October 19, 2011 5:58 PM
Magnus said (note the attribution),
"Steadily increasing phase error is to let there be a frequency error.
Frequency is the derivate of phase, so it comes with the territory.
So a 200 ns per second phase drift would provide a frequency error of 2
Hz on your 10 MHz. Can't have one without the other."
I understand, I'm just saying that if the absolute phase is wandering a bit
over tens of seconds, it's NOT an issue.
This is the famous "man who has two watches does not know what time it is"
problem.
Yup, I agree.. If I could count on the Tbolt output going dead if there's a problem, that would be wonderful, but my worst fear is that it would keep on going but be wrong enough to matter.
We are calibrating other equipment against the TB output to 1ppm so the TB would have to be pretty far off before it would matter, but I could see it happening.
With the two TBs running I can tell that "One of us is crazy", but I don't see an easy way for production people to know WHICH is crazy.
Out of curiosity, what would be the consequences of a steadily increasing
phase error? Would it offend your sense of perfection or would it have real
consequences?
If the frequency error is <<1PPM then I don't so much care. Beyond that it would be materially affecting our calibration.
My needs are simple, and the Tbolt's precision is the sort of "kill a fly with a sledgehammer" solution that I like.