Help me understand allan deviation measurements.

[Vgkid]

Cables came late yesterday,quick run.
did a comparision with the last measurement, just a png. Showing the results of 2 weeks of aging.

[DimitriP]

At least it's consistent ...

[Vgkid]

I modified the newer plot, and removed some of the peaks, from Tau from about 23s-_-50s they actually plotted extremely close to each other. almost perfectly overlapping.

[edpalmer42]

Now you're into the black arts of ADev measurements.  Add grounds, remove grounds, ground one end of the cable, ground the other end of the cable, try transformer coupling, try filters, amps, and attenuators, or pray to Dr. Allan to intercede on your behalf.  Actually he's still alive so that wouldn't work.

But it would probably be more productive to move on.  Your results aren't bad, they're just imperfect.  Welcome to the real world.  The only way to understand Allan Deviation measurements is to do them on as many different oscillators as possible, both good ones and bad ones so that you can see and learn about some of the odd things that oscillators can do.  As you test more oscillators you might find that your ripples go away.  "Oh, that's what caused them! Doh!"   

If you become a full-fledged Time Nut, you'll be on a quest to find the best oscillator(s) you can to use as references to compare to a new device under test.  You'll probably find that the best oscillator for short values of Tau (<~1000 sec.) is quartz.  A really good quartz might be good up to 10K sec, but those are few and far between.  For middle values of Tau (~1000 to 100K sec.) Rb will likely be best, but some models focus on low drift and don't care about ADev or phase noise.  With quartz in particular, there's a lot of variation from unit to unit even for the same model.  Two consecutive serial numbers might have drastically different performance.  Throw in the fact that one came straight from the factory and the other came from a Chinese recycler and all bets are off.

I've been surprised by the number of different areas that Time-Nuts stuff has led me into.  Programming, phase locked loops, Arduinos, construction tricks, new test equipment, multi-GHz RF things that I've never played with,  etc.  One of my Rb standards let me insert an RF probe into the cavity and look at the RF signals at and around 6.834 GHz.  I also measured the frequency response of the Cs tube in my Datum Cs frequency standard by varying the frequency around 9.192 GHz and measuring the 9 or 10 nanoamp output signal.

Have fun!

Ed

[Vgkid]

I tested another oscillator, A C-mac stp2145A. Since these are used, they require the pins to be cleaned, sometimes tinned/detinned before a reliable connection can be made. I would often get a frequency reading of around 11.68Mhz initially, but it would drop as the ocxo warmed up, but not enough. I did not save the file after I powered it on for 4 hours, but there was no ripple in the graph, but the frequency stability was starting in the 1E-9's. I will see if I can find the initial one.
I attached the 2 tim files as 1 zip file. Included are the original, and the modified version.

[edpalmer42]

When an OCXO wakes up after a long sleep, it can take a month or more to work the kinks out.  The rather ugly phase jumps in the original file could be exactly that.  During this initial period you'll also see aging that's so fast it will affect your measurements.  Many high-quality OCXOs state that aging can only be measured after 30, 60, even 90 days of continuous operation.  Now add in the fact that it's a used unit from China that was probably picked up by a big crane, dropped 5 or 10 meters into a bin, had other big, heavy things dropped on it, maybe shipped overseas in an open bin, and maybe removed from the board by heating the board and banging it on a table.  I'd be rather shaky too, if I endured that!  I'm amazed that these things work at all, never mind often work quite well!  Time will tell whether the crystal will ever settle down properly.

Did you notice the last part of the data that gives a nice straight line?  If you delete everything but that part, you'll find that the resulting ADev graph goes down into the low e-12 range!  That's looking very nice!  Both the frequency and phase plots are clean.  The drift shown on the frequency page is -5e-9 which is decent.  Hopefully, it will drop further as the unit runs.  Maybe this oscillator survived the recycling process intact.

11.68 MHz is completely wrong.  An OCXO should never be more than a few KHz off-frequency even if it's stone cold.  Use a scope to check the waveform and make sure that there isn't something terribly wrong.  For example, if the output is CMOS/TTL and the counter is set to DC triggering at zero volts you'll get some very strange results.  An internal fault in the oscillator that results in an extremely low output level could also cause gross errors because the counter is trying to dig through noise to find the signal.  I've seen that once or twice.

You mentioned that the frequency doesn't move far enough as the unit warms up.  Does that unit have EFC capability?  If so, you might need to hook it up to a pot to get the frequency correct.  If the unit is old, it may have drifted far enough off frequency that it can't be corrected.  I have cracked open a few oscillators to modify them to bring them back on-frequency.  It's a lot of work unless it's a really, really good oscillator or just as a learning exercise.  If an off-frequency oscillator has really good ADev, it's still useful as a reference to compare other oscillators to.  The frequency error isn't important there.  In fact, there are other ways to measure ADev that require an off-frequency oscillator.  Another interesting fact is that if the frequency starts high and drops as the unit warms up, it's probably an AT-cut crystal.  If it starts low and rises, it's probably an SC-cut crystal.  There are other crystal cuts, but those two are by far the most common.

Here's a TimeLab hint for you.  While you're in the middle of a data run, you can't edit the data.  But you can save the .TIM file without affecting the data collection.  Now you can reload the .TIM file without affecting the data collection.  Now you can edit the saved data as much as you want.  If you're doing a long data run, it's not a bad idea to save the .TIM file occasionally in case Murphy sticks his nose in where you don't want it!

Ed

[Vgkid]

Should have clarified some stuff. For the 11.68MHz(I think it was 10.68,still far off) the frequency actually dropped as it warmed up, but not much. Unfortunately I scoped it and it did not show any strange abnormalities in the waveform, but even when using the counter in the scope it was still off.  I had actually noticed this issue with the C-mac CFP0-4's, and the STP2145A's. The CFP0-4's are actually (*most likely) NOS. *I bought 50 of them, and at least 30 have sequential part numbers.  Also the boxes that they came in were marked with C-Mac, and had evidence where the original shipping labels were taped on. The worst part is that some tape was placed over the shipping labels(so you can't read them), or even remove the tape.
      Back to the frequency issue.   
   I noticed when I started testing the original box of CFP 's  I noticed that they would often settle to 10.68M when I soldered wires to the ocxo. So I desoldered them, and grabbed another. I did this with 3 separate times, same issue. I grabbed another, all to the same effect. In frustration I desoldered the VCC/ground leads from the last ocxo under test. Grabbed a pair of spring clips and it worked. For fun I tried to “kill” one of the incorrect ocxo's. I cranked the voltage up to 30+V...and it worked, the ocxo that is. So I dialed the psu back to 12v, and it kept running. I marked that one, as to not use it. Skipping forward a few months. This time using the old CFP(the one used originally in this thread) I had the same issue as earlier, and switched to using spring clips. I eventually soldered it in so that it worked. Same thing for the new CFP that is currently running. A lot of fiddling was needed to get the correct output frequency. * The connections to VCC/Ground were solid.
      Onto the STP2145A
   When It came to powering up the ocxo the same issue had arisen, and I went through the same steps. This time I will document what it took to remedy this. When I first started using this for this current test. I cleaned the terminals with a cotton pad soaked in IPA. Soldered on he power connectors. Cleaned those off with IPA, and powered up the unit. Same thing, so I desoldered the wires(did not remove the excess solder), and attached some spring clips. Same large frequency offset. I grabbed my spare 2145A and powered it with spring clips, it worked. Perplexed I completely removed the solder(with braid) from the power pins. And cleaned those off, then resoldered the power wires. It worked this time. The strangest thing about this is that the ocxo drew a reasonable amount of current, it even dropped off as the ocxo warmed up. The voltage to the VCC pin was the same as the output of the psu, sometimes connection issues cause strange results.
   End novel.

[edpalmer42]

I can't believe all the weird problems you've been having.  Did all the members of the Murphy clan move into your neighborhood??  Soldered joints don't cause frequency errors!  Clipleads don't give better connections!  This is just nuts!  You saw the sloppy wiring I did a few pages back and I still had better measurements than you're getting.

I'm starting to wonder if there's something in your lab or nearby that's injecting noise into your circuitry.  Maybe noise on the AC ground or RF noise being coupled into your circuits.  Maybe a noisy switching power supply.  Could be a cracked solder joint in a ground connection.  Here are a few increasingly desperate ideas for you to try. 

I think your 5335A counter can do this.  Take a signal that you can do a clean connection to.  BNC to BNC - double-shielded if possible.  NO CLIPLEADS!!!  Measure the period on Channel A with statistics.  Do the same on Channel B.  Is the Standard Deviation similar on both channels?  Yes, I know your counter agreed with your scope, but this is an easy test to do.  You might want to do it to your scope as well, just to be sure.  Your earlier test of the counter's noise floor was good, but I don't think it would pick up a bad ground on one channel.

I think you said that you're collecting the Timelab data on a laptop.  Shut it off, disconnect it from your counter and unplug the power brick from the wall.  Redo the Standard Deviation test.  Any change?

Make sure that everything is plugged in to the same power source.  Turn off everything you can that's not part of the measurement.  This includes things like fluorescent or LED lights.

You said that you don't have a spectrum analyzer.  Any chance you can borrow one to look for noise sources?  Could you borrow a different counter, preferably a different model?

I made myself a poor-man's Faraday Cage out of a tin-plated steel biscuit tin that measured about 22x22x9 cm.  I mounted both regular and isolated BNC connectors and added some holes with metal grommets for feeding through wires.  The grommets prevent the sharp edge of the hole from cutting the wire insulation.  Feedthrough capacitors would have been better, but I didn't have any handy.  It works surprisingly well at attenuating RF.  Be sure to sand off any paint or laquer that might prevent the top from making good contact with the body.

This is really starting to bug me!! 

Annoyed Ed 

[edpalmer42]

Hey, all you lurkers out there, if you have any ideas that might help Vgkid, let's hear them!

Ed

[DimitriP]

Hey, all you lurkers out there, if you have any ideas that might help Vgkid, let's hear them!

Ed

Ha...I was wondering when you were going to get "lonely"...
I'd like to see a couple of pictures of the setup. Close up and overall.
It does sound like Murphy lives next door....

[edpalmer42]

Ha...I was wondering when you were going to get "lonely"...
I'd like to see a couple of pictures of the setup. Close up and overall.
It does sound like Murphy lives next door....

I didn't really think an invitation was necessary.  The more, the merrier!  With over 2800 views on this thread there's got to be at least a few good ideas!

Ed

[DimitriP]

Here are a few increasingly desperate ideas for you to try. 

Checking STD DEV of the counter's own timebase vs the oscillator in the "lab position" with "everything else" "normal" and then turned off (laptop, cell phone, tablet,  soldering station,oscilloscope overhead lamp , whatever else is powered on , etc etc etc and then (especially if the numbers are close)  taking the counter and oscillator in another area  altogether and checking them again should reveal (hopefully) a dramatic difference in behavior 

[DimitriP]

Ha...I was wondering when you were going to get "lonely"...
I'd like to see a couple of pictures of the setup. Close up and overall.
It does sound like Murphy lives next door....

I didn't really think an invitation was necessary.  The more, the merrier!  With over 2800 views on this thread there's got to be at least a few good ideas!

Ed

You must have this thread confused with one about probes, soldering stations, soldering tips , which DMM or which oscilloscope to buy

[Vgkid]

I will provide pictures later today.
Ironically I actually ran one of the test(with the old ocxo) with nothing powered on in the room. The only things that were powered on were the psu's, the 5335A, and the laptop was running on battery(eco mode, screen closed,wifi off), same ripples in the graph. Though I initially plotted the stp over a short run(90min) only after it was powered on 2hr and no ripples were present. I did manage to find it.

[Vgkid]

Here is the file.

[DimitriP]

I did manage to find it.

I thought I was following along, then I got lost.

What's the "it" you managed to find? The source of the wiggles?
What was causing them?

[Vgkid]

'It' was the timelab file that did not have any wiggled in it. Rather strange since the 28145asian had only been powered on 12 hours. Even though the sample run was 90min. The other runs had wiggles in them. I'm currently returning the test at the 90min sample run, to see if it still fails to produce those wiggles.

[Vgkid]

As requested here are the pics.
A overview of the setup. The soldering iron is(has been) powered off for the duration of the test, along with everything else that is not essential on the bench.

The first image shows what is in the plastic box. A STP2145A ocxo, running a (2) 10K(WW,%.05,??TC) resistor divider. The 2 thin leads are a twisted pair leading to my HP 3456A measuring the output of the Vref. It is rather stable over a day, though not perfect. The ocxo connects to a counter(50ohm) using a probe with alligator hooks.

The second image shows a close up of the whole setup. it is not very pretty.

The third image shows the CFP04 oscillator.* A cardbpard box actually covers the unit, and yes it is sitting on a box of ocxo's.
A resistor voltage divider sets the output frequency. It is composed of (2) 11.8K(WW,.01%,TC5ppm), The output connects to the scope through a piece of RG58cable.

A close up of the setup, the abundance of electrical tape is used to keep anything from shorting out. yes I did clean all terminals after applying the tape to remove any residue.

[DimitriP]

Since you are still in a "war of the wiggles", next time before testing 
A) I'd eliminate monitoring Vref so there are no long leads hanging off of it.
B) I'd twist the twisted pairs much more tightly up to the point of connection so you only have an inch or so "straight"
C) or I would give up on the twisted pair and use coax for power as well
D) I'd  add a .01, 0.1 decoupling caps on the Vefc and Vin as well
E) Does the switch of the soldering iton switch the primary or the seconday off?.... I'd unplug it altogether.
F) and I would unplug from the wall whatever other wall warts may be plugged in for good measure.

Last thought just popped in before I hit enter...how about using one of the already tested ocxos as the reference for the counter? Hmmm I'll hit <Enter>  now

[edpalmer42]

I agree with Dimitri's comments, although you shouldn't have to use coax for the power feeds.

Have you always had the EFC twisted pair attached?  I strongly agree with Dimitri that it should be removed.  The EFC input is high impedance and is very sensitive.  Any noise on the EFC line will cause FM modulation on the oscillator's output.  This is identical to jitter which equals bad ADev.  Make this the first change.

I'm assuming that the (scope?) probe is x1.  If it was x10, the attenuation would be brutal when used on a 50 ohm input.  I've got a couple of 100 MHz probes that have a resistance in the 100 - 300 ohm range when switched to x1.  Trivial when used with a 1 Mohm scope input, but more significant when used with a 50 ohm counter.  It would also mess up the 50 ohm load on the oscillator.  I don't know what the frequency response of a x1 probe would be when used on a 50 ohm input.  Move the probe from the counter to a 50 ohm terminated scope input and check the level that the counter will actually see.

I like the way you isolated the units from drafts with the cardboard and plastic boxes.  I also like the way you connected the coax onto the CFP04 oscillator.  I was thinking about the best way to do that and came up with the method you used.    If it looks like the scope probe might be causing problems, replace it with another piece of coax.

Ed

[DimitriP]

I'm assuming that the (scope?) probe is x1.  If it was x10, the attenuation would be brutal when used on a 50 ohm input.  I've got a couple of 100 MHz probes that have a resistance in the 100 - 300 ohm range when switched to x1.  Trivial when used with a 1 Mohm scope input, but more significant when used with a 50 ohm counter.

You must mean X1 in the italicized sentence above? Was this a reading test ?

[edpalmer42]

I'm assuming that the (scope?) probe is x1.  If it was x10, the attenuation would be brutal when used on a 50 ohm input.  I've got a couple of 100 MHz probes that have a resistance in the 100 - 300 ohm range when switched to x1.  Trivial when used with a 1 Mohm scope input, but more significant when used with a 50 ohm counter.

You must mean X1 in the italicized sentence above? Was this a reading test ?

No, I meant what I said.  A x10 scope probe expects to be used with a 1 Mohm input so it has ~9 Mohm in series.  If it was used with a 50 ohm input, the attenuation would be ~50 to 9M or x180K instead of x10.  I think that qualifies as 'brutal'........or maybe 'lethal' is a better word. 

Ed

[Vgkid]

The probe is in the 1X setting. I was measuring the efc outputvoltage with 3456A. Over 2 days the std dev was around .1E-9, while the V-min/max was less than 60 counts. It is around 8.0v, with center voltqge around 4.2. I think i could do better.

[edpalmer42]

I was measuring the efc outputvoltage with 3456A. Over 2 days the std dev was around .1E-9, while the V-min/max was less than 60 counts. It is around 8.0v, with center voltqge around 4.2. I think i could do better.

It's very unusual to monitor the EFC voltage directly because you're almost guaranteed to degrade the oscillator's performance by doing so.   The correct way is to disconnect the monitoring leads and then check the oscillator's output.  If it looks like there's a problem with the EFC, make changes and then see if the performance of the output has improved.

You're wasting time by trying to measure the EFC voltage directly.

Ed

[Vgkid]

The efc test were done initially. For the stp's A-dev test, the wires were unsoldered from the pins/removed. The cfp04 didnt have any.
Last night I tested the cfp04, at least it is consistent.
grn-1day(,maybes less,not noted),pink-4wk,blue-6wk.