USA calibration club

[orin]

Somewhere back in this thread I thought the SVT got measured with a far better meter than I've got, and it was very close. Though I have three Fluke 731s that I can intercompare, it's been more than 13 years since the last traceable cal on my best one. I call that the "golden" unit because I modified it for near zero TC and it's the one that gets sent out. They're good, but I'd be astounded if they were that good. In the years since getting a cal, I've compared to various new Agilent 6.5 digit meters purchased at the day job. I'm always within a couple ppm, but I don't know how good those meter were when received- the chicken and egg problem!

FWIW, my comparison of my SVR-T and my Fluke 731B is here: https://www.eevblog.com/forum/testgear/geller-labs-svr-and-svr-t/msg282582/#msg282582

Orin.

[Conrad Hoffman]

Thanks, your graph and my results show the same signature wander, albeit very small. I need to do the same test between my different 731s, but the SVT is definitely on a similar level of performance.

edit- unit is now in the hands of the USPS and on the way to Dr. Diesel.

[sokoloff]

OK, this is officially bad. I spent most of Independence Day farting around with old lab equipment (and doing a whole lot of yak shaving), but I got myself setup to the point where I have GPIB interfaced equipment working! 

The reason this is officially bad, is it's now caused me to wonder, "Which of my 6.5 digit meters is more accurate?", "What does it mean to 'calibrate' a meter?", "Why wouldn't my meters all read 0.0 microvolts DC with shorted leads?", and a whole host of other gateway-to-madness questions...

[Conrad Hoffman]

I spent some of the day tweaking my two other references to match my "golden" unit, and then doing data collection between two of them. Things are good, sub ppm wander, but now I need to know the cause of it. Thermal issues with the leads, leakage currents in the measurement setup, temperature variations, the dehumidifier turning on and off, with the associated air currents from the fan, HF interference from the LED lights in the ceiling and on and on. Also started improving my GPIB logging program, since this was the first time I actually used it. I neglected to account for the midnight rollover on the data timestamp, plus I should display the measurement count so I have some idea how far along a given run is. Should probably fix the abort button as well. It will be great when the SVT hits the cal lab so we can all see where we really are.

BTW, I have a couple HP3455A 6.5 digit meters. The one I use regularly works as expected, zeros correctly and reads the same for both polarities. The one on the shelf meets all specs, but it has a slight polarity difference. As for accuracy, they're old and don't have very sophisticated references, but will still hold 1-2 ppm after warm-up for years at a time. In general, you should be able to adjust almost any 6.5 digit meter to near perfection, it's just a matter of how long and under what conditions it stays that way. 

[dr.diesel]

edit- unit is now in the hands of the USPS and on the way to Dr. Diesel.

This might just work out, if my schedule holds my two 3458A's will be at @CalMachine's this Thursday.

[vindoline]

In general, you should be able to adjust almost any 6.5 digit meter to near perfection

Oh dear, now you've said it....

[CatalinaWOW]

OK, this is officially bad. I spent most of Independence Day farting around with old lab equipment (and doing a whole lot of yak shaving), but I got myself setup to the point where I have GPIB interfaced equipment working! 

The reason this is officially bad, is it's now caused me to wonder, "Which of my 6.5 digit meters is more accurate?", "What does it mean to 'calibrate' a meter?", "Why wouldn't my meters all read 0.0 microvolts DC with shorted leads?", and a whole host of other gateway-to-madness questions...

Or another variant - if you have two meters that track each other to 10 microvolts over time and temperature, and another that has a constant difference (All same make, different models), which is right and does the tracking indicate that they are really stable, or just have the same kind of errors?

I understand exactly.  This is where I am also.  After spending a few days logging my crude standards and calibrating their sensitivity to supply voltage and temperature I am starting to worry about some of that real volt nut stuff.  I started out thinking that knowing to 100 microvolts would make me happy.  I now think I can know within 10 microvolts what the standard is saying when it arrives and am working my way down to one and maybe even a third.  I think I have a path, but don't know if I am really that deep in the disease.

[alm]

BTW, I have a couple HP3455A 6.5 digit meters. The one I use regularly works as expected, zeros correctly and reads the same for both polarities. The one on the shelf meets all specs, but it has a slight polarity difference. As for accuracy, they're old and don't have very sophisticated references, but will still hold 1-2 ppm after warm-up for years at a time.

Early 3455A used a selected LMx99, just like 90% of the other 6.5 digit meters out there. Only the proprietary selection process may be different (I do not know the p/n they used, since they considered the reference module as not field repairable). So no reason to expect the stability of their reference to be any worse than more modern meters. Of course the supporting/input/ADC circuit may be different, and may also affect drift.

Funnily, they replaced the LMx99 with a custom heated zener in later 3455A units, which (or a very similar part) was also used in early 3456A units, but subsequently replaced by a selected LMx99 again in later 3456A units. Not sure what the story behind that was. I do know that they retroactively tightened the spec of the 3456A after they gained more confidence in their reference, so maybe the switching between LMx99 and the custom reference was also related to long term stability data collected over years that proved the LMx99 to be more stable than their custom zener.

[dr.diesel]

Look what just showed up.

I'm gone till Monday, but will get right on it when I get back.

[cellularmitosis]

Woohoo!

[orin]

BTW, I have a couple HP3455A 6.5 digit meters. The one I use regularly works as expected, zeros correctly and reads the same for both polarities. The one on the shelf meets all specs, but it has a slight polarity difference. As for accuracy, they're old and don't have very sophisticated references, but will still hold 1-2 ppm after warm-up for years at a time. In general, you should be able to adjust almost any 6.5 digit meter to near perfection, it's just a matter of how long and under what conditions it stays that way. 

3455As and AD587 (Geller SVR(-T)s) don't get along well if you reverse the polarity unless you put a pretty big capacitor on the SVR output or are very careful about shielding and guarding.

I would see 9.9998X with the polarity normal and -9.989XX with the polarity reversed (I don't recall the X digits, they depended on temperature drift in the 'lab').  I saw this on two different 3455As.  I did not see it on a 3456A.

I could see 120mV pk-pk at 570kHz when I looked at the signal with a 10X scope probe, so I suspected AM radio pickup.

It took 0.33 uF on the SVR output to make the 3455A read the same magnitude with the polarity reversed.

The guarding alternative was to use shielded twisted pair from the power supply to the SVR and from the SVR to the 3455A (shield grounded at the power supply, connected straight through at the SVR and connected to the Guard terminal at the 3455A).  This got the difference in magnitude down to 15 counts.  It still needed 4.7 nF capacitance to get the magnitudes within one count with the polarity reversed.

[Conrad Hoffman]

I didn't see any particular problem with the SVT and my 3455A- the results agreed with the Fluke 845A and the 3478A and the three 731s, just with less resolution. The last digit of the 3455A will wander a couple counts. I don't have the data logging set up for the 3455A, but that would be the smart thing to do. Too many projects, too little time. The problem with my other 3455A seems to occur with any voltage source, so I'm assuming it's inherent in the unit. It's not a huge amount, just a few counts, but too much for my OCD. I don't remember there being any adjustment to trim it out, so one might have to select components or something.

The whole shielding/guarding thing has pitfalls. That's why I used a wall wart and the regulator board to power the SVT, as I don't know the ground isolation of my big bench supplies. It's the current paths you don't know that will throw things off. On that note, last night I did a run comparing two of my 731s. A couple periods of maybe 5-15 minutes had 1/4 ppm steps I need to track down. It could have been air currents on the connections, or even popcorn type noise in a reference. The trials and tribulations of a voltnutter.

[alm]

3455As and AD587 (Geller SVR(-T)s) don't get along well if you reverse the polarity unless you put a pretty big capacitor on the SVR output or are very careful about shielding and guarding.

My theory would be the higher capacitance from low terminal to ground due to the guard being connected to low by default would cause this issue. But disconnecting guard from low and connecting a separate wire from guard to SVR ground should fix that. Sounds like you had to go through much more trouble. Another option is the 3455 injecting current into the circuit. I remember Bob Pease complaining about that, so it could definitively have been the 3455. Adding some large-ish values series resistors might help and should have a negligible effect on the voltage measurement.

The last digit of the 3455A will wander a couple counts. I don't have the data logging set up for the 3455A, but that would be the smart thing to do. Too many projects, too little time. The problem with my other 3455A seems to occur with any voltage source, so I'm assuming it's inherent in the unit. It's not a huge amount, just a few counts, but too much for my OCD. I don't remember there being any adjustment to trim it out, so one might have to select components or something.

The service manual has a procedure for isolating DCV noise. Unlike modern manuals, it tells you to use a known good board, but subsequently explains how to do the tests if you do no have access to a known good board. This thread suggests that it might be dielectric absorption in the integration capacitor. Could well be an aging issue. The 3455 is much more sensitive to that than the newer meters. It is one issue that they explicitly fixed in the 3456 design. No argument that investigating it would take time .

[orin]

The two or three digit wander* in Hi-Res mode with Auto-Cal on the 3455A is due to dielectric absorption in the integrating capacitor.  I suspect it was better when the meter/capacitor was new.  Other than the (physically huge) Russian teflon capacitors that can be found on ebay, I've not found anything better than the original capacitor.  With a couple of 39nF teflon capacitors in parallel instead of the original 82nF polypro, my last digit jitter is around one count.

*EDIT:  This wander is in time with the clicking of the auto-cal relays.

[Conrad Hoffman]

Is the HP3455A wander due to dielectric absorption only present when measuring a voltage? No voltage = no voltage on the cap and nothing to absorb?

My HP3478A is particularly noisy on the 30 mV range- can anybody tell me what's normal with a shorted input? The manual on that one does suggest turning off the auto-zero if change injection into the DUT is a problem.

[zhtoor]

Is the HP3455A wander due to dielectric absorption only present when measuring a voltage? No voltage = no voltage on the cap and nothing to absorb?

My HP3478A is particularly noisy on the 30 mV range- can anybody tell me what's normal with a shorted input? The manual on that one does suggest turning off the auto-zero if change injection into the DUT is a problem.

i have a HP3468A, which with inputs shorted is +- 1uv (Least significant digit goes 1/2/1/2/1/2 .....) on 30mv range.

regards.

[alm]

The 3468A has a 30 mV range?

[zhtoor]

The 3468A has a 30 mV range?

sorry, my *bad*, 300mv. 

regards.

[orin]

Is the HP3455A wander due to dielectric absorption only present when measuring a voltage? No voltage = no voltage on the cap and nothing to absorb?

It's present when you have Auto-Cal on as the meter is continually measuring a set of references which include a +10V reference and a -10V reference.

So when the meter measures the +10V reference then your input, your input reads a count or two high.

When the meter measures the -10V reference then your input, your input reads a count or two low.

If you log the readings, you can see a pattern as in the attached graph.  You'll see the same pattern (based about 0V) if you short the input.

If you turn Auto-Cal off, you'll just see at most one count jitter, but it will drift badly.

[Conrad Hoffman]

I must have the best 3455A ever made, or it's broken. It wanders by about 1 count after a good warm-up. I've never paid much attention to what it does with auto cal off.

[cellularmitosis]

I just went back through the posts in this thread and made an index: https://goo.gl/GHyj67

I'll try to keep it updated as we go along!

[cellularmitosis]

While waiting on my GPIB "spy" board and logic analyzers to arrive, I got my 34401A logging via RS-232 this evening.

It turns out you can put it into "talk only" mode, which makes it super easy to get started logging data from this meter.  See my post here: https://www.eevblog.com/forum/testgear/34401a-interface-to-pc-rs232-or-gpib/msg1257501/#msg1257501

[dr.diesel]

Finally some data!  I was unable to get extended runs without power outages due the storm pounding the midwest has seen over the past week.  I've attached two graphs (LibreOffice Graphing is a huge PITA, any other Linux suggestions?) one capturing the entirety of the spikes and one narrowed down a bit.  I need to track down what is causing the spikes, something enviromental as both 3458As see it, might be my shop air compressor.  Note I did not offset the the two readings for graph readability and both are back from a fresh cal.  I utilized the included PS in a consistency attempt, fed via an old Heath linear PS set at 20V.

Cabling was solid twisted ethernet cable, utilizing the 3458A screw terminal, no banana.  Connection at the DUT was copper Mueller Electric Co BU-34C, crimped, not soldered, which I'll forward on when I ship the reference to the next participant. 

[dr.diesel]

Forgot to mention, 3458A setup, from the PI lognut scripts, hacked to read both 3458As and a Maxim ds18b20 temp probe.

# Setup HP 3458A
inst.write("PRESET NORM")
inst.write("OFORMAT ASCII")
inst.write("DCV 10")
inst.write("TARM HOLD")
inst.write("TRIG AUTO")
inst.write("NPLC 200")
inst.write("AZERO ON")
inst.write("LFILTER ON")
inst.write("NRDGS 1,AUTO")
inst.write("MEM OFF")
inst.write("END ALWAYS")
inst.write("NDIG 9")
inst.write("DELAY 0")

[CalMachine]

Finally some data!  I was unable to get extended runs without power outages due the storm pounding the midwest has seen over the past week.  I've attached two graphs (LibreOffice Graphing is a huge PITA, any other Linux suggestions?) one capturing the entirety of the spikes and one narrowed down a bit.  I need to track down what is causing the spikes, something enviromental as both 3458As see it, might be my shop air compressor.  Note I did not offset the the two readings for graph readability and both are back from a fresh cal.  I utilized the included PS in a consistency attempt, fed via an old Heath linear PS set at 20V.

Cabling was solid twisted ethernet cable, utilizing the 3458A screw terminal, no banana.  Connection at the DUT was copper Mueller Electric Co BU-34C, crimped, not soldered, which I'll forward on when I ship the reference to the next participant.

Do you have your 3458As on top of each other?