Help me understand allan deviation measurements.

[Vgkid]

^^^
Better than hearing a voice that answers you,when you are alone..(my 1 ghost experience)
ended up unplugging everything from the bench. As i plugged in each instrument, i measured the voltage from the chassis to the bench everything reads 60v

[Vgkid]

Moved the bench so it is not against the baseboard heater. Now everything is at 15vrms
:/

[DimitriP]

Common ground is not that common sometimes...
Was this heater in use when you were taking measurements ?

[Vgkid]

Over the whole time, no.

[Vgkid]

Progress is made...
Not into the upper 7x10^-12, like the last one I ran, a few days ago(had oscillations in it).

[Vgkid]

Oscillations are still present, atleast they are reduced, as compared to previous measurements. Still trying to figure out why all of the gear is floating at 15v, beats 60v...
Attached the TL file, the finished one was run over 2 days, 24hr. The other was run over 6 hours last nght.
On the 24 hour run, what would make the graph look like this, as opposed to the usual AlDev graph?

[edpalmer42]

What is it about the 24 hour graph that looks wrong to you?

Ed

[Vgkid]

So if i allowed the counter to count for 1more hour(allowing my to remove the 3 spurious freq jumps)/ drew a line following the curve. Would this allow me to approximate the 24 hour drift rate?

[edpalmer42]

So if i allowed the counter to count for 1more hour(allowing my to remove the 3 spurious freq jumps)/ drew a line following the curve. Would this allow me to approximate the 24 hour drift rate?

You need to run longer than 24 hours to get a 24 hour Allan Deviation value.  In Timelab, if you use Ctrl-E to turn on the error bars for the Allan Deviation screen, you'll see that the end of the graph has really wide error limits.  So you might have to run your test for 2 or 3 days before the one day value is reasonably well defined.

Ed

[Vgkid]

Ok, might be a while.
Equipment voltages are down to around 1v wrt the bench.

[Vgkid]

So I ran the CFP04, and the STP2145A for 48 hours. The CFP04 had only ben powered on almost 1 month, the STp2145A has been powered on for several months.

[edpalmer42]

The cfp04 graph looks okay.  If you had run the test longer, the lowest part at 1000 - 2000 sec. wouldn't have changed, but the hook on the right hand would have disappeared, the graph would have risen smoothly from the lowest readings, and the 1 day results would have been 1 or 2 e-10.  A reasonable result that shows some low tau limitations in your measurement system.

The 2145A graph is somewhat suspicious.  The AlDev goes out to almost 80K seconds and it doesn't show any aging yet.  The phase plot shows a frequency change at ~121K seconds, but that hasn't yet appeared in the AlDev results.  If I remove the data after the frequency change, things get even more suspicious.  Are these measurements time interval measurements versus the LPRO or frequency measurements?

Ed

[Vgkid]

Frequency measurement, I can not seem to get the TI mode to give me anything other than negatively sloped line. I do want to see if i can arm the counter from a 1pps signal.

[edpalmer42]

Frequency measurement, I can not seem to get the TI mode to give me anything other than negatively sloped line. I do want to see if i can arm the counter from a 1pps signal.

That's suspicious.  If you measure the time interval between two oscillators at slightly different frequencies, the result should gradually drift through all values until it reaches an interval equal to the period of the frequency involved and then wrap back to zero.  The numbers could increase or decrease depending on which oscillator is faster or slower.  On a graph it will look like a sawtooth.  e.g. for two 10 MHz signals, the result should go from 0 to 100 ns.  If the frequency difference between the sources is too high, you won't see the sawtooth, it will just look like random numbers in the same range.  If your counter can't do that, it makes me think there's something wrong with your counter which casts suspicion on all your results.

I know that earlier in this thread, KE5FX recommended frequency measurement.  I've always preferred time interval because some counters work much better in that mode.  My HP 5370B and Wavecrest DTS-2077 are like that.  Also, with time interval, you can be sure that you're only measuring one interval and aren't doing any averaging which will artificially improve your results.  I think that's what's happening in your recent results.

Ideally, the start signal would be 1 PPS and the stop signal would be whatever the DUT (Device Under Test) provided.  Configure the counter to use its internal reference.  It should be able to take the reading and reset before the next pulse so deadtime shouldn't be an issue.

I'd recommend that you take another look at time interval mode.  Timelab understands phase wraps and does a pretty good job of unwrapping them.

Ed

[edpalmer42]

Hi Vgkid,

I just made an interesting discovery that I wanted to pass on to you.

I always wondered why the slope on your graphs was so shallow.  All my graphs start out at the counter's resolution and drop 1 decade for every decade of time.  Yours are more like one decade drop in two decades of time.  I recently decided to measure frequency and use it for AlDev calculations.  I don't think I've ever done that.  Guess what?  Same shallow slope.

The problem is that since it takes so long for the shallow slope to get down to the e-12 area, the OCXO has already passed its lowest value and is starting back up due to aging.  It also seems like it has a relatively high noise floor so that further distorts the measurements.  I think that's why your results haven't been particularly good.

This doesn't explain your oscillations, but it's still useful info to have.

Ed

[Vgkid]

@ Ed, thanks for that piece of information. I have always wondered about the shallowness of my slopes, when looking at others al-dev graphs.

[awallin]

I always wondered why the slope on your graphs was so shallow.  All my graphs start out at the counter's resolution and drop 1 decade for every decade of time.  Yours are more like one decade drop in two decades of time.  I recently decided to measure frequency and use it for AlDev calculations.  I don't think I've ever done that.  Guess what?  Same shallow slope.
The problem is that since it takes so long for the shallow slope to get down to the e-12 area, the OCXO has already passed its lowest value and is starting back up due to aging.  It also seems like it has a relatively high noise floor so that further distorts the measurements.  I think that's why your results haven't been particularly good.

fwiw in frequency counter mode I made noise-floor measurements of a 53230a counter:
http://www.anderswallin.net/wp-content/uploads/2015/06/RCON_vs_CONT.png

by default it uses a CONT mode which has a lot of internal averaging and a 1/sqrt(tau) slope. calculating ADEVs from this data is questionable.
The undocumented RCON mode looks more like a time-interval measurement noise floor and the slope is 1/tau.

not sure if the counting mode is adjustable on other counters. on new computerized counters it should be possible.

[Vgkid]

Mini update: no fix on  the allen deviation graph. The main cfpo4 has still been kept under power, though it was powered off for about a week. Previous to being de-powered it has been behaving well.  Though it will be depowered(along with everything else) for a week.
Regarding the ocxo wrong frequency turn on bug, I tested 14 cfp04's,  and they did not display this glitch.
 In related news, I managed to break one of the ocxo's. Teardown if you want it.

[edpalmer42]

A while ago I was making some measurements and direct cable connections weren't appropriate so I used a scope probe for one channel of the TIC.  The results were trash - I had lots of wiggles.  I was going to investigate it more before posting it, but never got around to it.   

Are you still using a probe on one channel?

Ed

[Vgkid]

Direct connection for frequency. I do need to pick up come bnc cables to do some bulk testing(much much later).