Guildline 4410 or Advantest R6581/T For Lab Cornerstone Standard?

[bdunham7]

On page 19 of the HP Journal issue introducing the 3457A they describe how they have come up with a more accurate drift figure for the LMx99 reference, and how the datasheet for the 3456A has been amended to reflect this. I have not found this datasheet, but on page 192 of the 1987 catalog (warning, large PDF) they give a 23ppm + 2 counts figure for 1y accuracy on 10 VDC and 15 ppm + 2 digits for 90 days for the 3456A. I attached the relevant table.

I believe this is better than the 3457A which did use close case calibration and laser-trimmed resistors, so I guess their engineers apparently disagreed that the old way of constructing was less stable. But obviously they weren't specifying decades-old instruments.

That's an interesting read.  One takeaway point is that they apparently greatly improved their reference characterization between 1984 and 1987.  Would those new specifications apply to older units that had been built before they implemented this method?  Is there a known serial number breakpoint for this change?  Fluke also extended the specs and cal intervals on their 850xA series IIRC so I'm guessing there was some competitive pressure.  I imagine customers would have really pushed for a 1-year cal interval.

I don't think you can or should directly compare specs between different manufacturers, different eras or even the 3456A and 3457A (due to the ranging differences).  Do you know of any explicitly stated or implicit confidence intervals attached to these specifications?  I've never seen them for these older instruments.  I've rehabbed a fair number and my experience is there's a significant difference between older and newer (A-brand) meters of comparable specifications in both as-found tolerances (if the calibration data or settings have not been disturbed) and ability to meet/exceed the published specifications once I've done my best to restore them.  At the very least I would say that the actual performance of newer models from HPAK, Fluke or Keithley will be at least as good as their elder counterparts, even ones that have slightly better published specs. YMMV, that's just my experience.

A null meter like Keithley 155 only makes sense if you are adjusting your device (or a KVD) to equal the standard. They are optimized for measuring if a voltage is null, not how far a voltage is away from null. Generally in metrology standards have a value that is closeish to the nominal value but they are not adjusted to have exactly that value, so the voltage standard might get assigned a value of 9.9938V. So if you want to adjust your device to be as close to 10V as possible, you want it to be about 600uV higher. A Keithley 155 is not very useful for this, since it will just be an analog meter with mediocre accuracy. I'd rather use a good bench DMM or if you want to go crazy a nanovoltmeter for this.

I would agree with you if the reference was actually 60ppm from nominal, but aren't Fluke 732A/B references adjusted much closer that that?  I have no experience with those and the only examples I could think of were the ones on xDevs--but that info is not available at the moment.   If it is within 10ppm or 100µV, then even a 5% error on the null meter would only result in a 0.5ppm error or less.  Measuring <100µV with a DMM has its own challenges, not that I haven't done it myself.

[alm]

That's an interesting read.  One takeaway point is that they apparently greatly improved their reference characterization between 1984 and 1987.  Would those new specifications apply to older units that had been built before they implemented this method?  Is there a known serial number breakpoint for this change?

I haven't found more details like a serial number breakpoint or anything like that. My interpretation is that the change is mostly based on them having collected more data on the reference over the past years, so would largely apply retro-actively. But I couldn't find any more details, so I'm not sure.

I don't think you can or should directly compare specs between different manufacturers, different eras or even the 3456A and 3457A (due to the ranging differences).

For the ranging differences, you can just do the math for any measurement, like 10V, and calculate the uncertainty from either meter. And given the unfavorable ranges of the 3457A for decade values common in metrology applications, it will only get worse for the 3457. In metrology the instrument is assumed to perform within factory specifications unless proven otherwise (either way). So a meter specified for 20 ppm/year can be assumed to stay within 20 ppm for that year, while a 30 ppm/year meter that actually performs the same must be assumed to have 30 ppm uncertainty over that year unless you an prove with multiple years of data that real world performance is better. So specifications are important, and I find it hard to believe manufacturers would not tighten those specifications if they could.

Do you know of any explicitly stated or implicit confidence intervals attached to these specifications?  I've never seen them for these older instruments.

The default (also specified as such by the the Guide to the expression of uncertainty in measurement, GUM) is to assume the specifications are hard limits and the actual value can be anywhere within that limit (rectangular distribution). For example 0.1% resistors are normally tested to be within 0.1%, it may not be a normal distribution, but there is a sharp cut-off around +/- 0.1%. Those are the easiest to give by manufacturers. Sometimes manufacturers give more specific specifications with a 95% or 99% confidence interval, which allows the customer to use more optimistic uncertainty figures, but this is the exception rather than the rule. In all other cases a conservative interpretation is used, which leads to larger uncertainties. Of course in the case of DMMs there is likely an underlying student-t distribution or something like that that is much narrower than the limits, and the majority will probably be not so close to the limits, but since they don't tell people assume the most conservative interpretation. This Fluke white paper touches on those concepts.

I would agree with you if the reference was actually 60ppm from nominal, but aren't Fluke 732A/B references adjusted much closer that that?  I have no experience with those and the only examples I could think of were the ones on xDevs--but that info is not available at the moment.   If it is within 10ppm or 100µV, then even a 5% error on the null meter would only result in a 0.5ppm error or less.  Measuring <100µV with a DMM has its own challenges, not that I haven't done it myself.

Voltage standards are not normally adjusted after leaving the factory. Adjusting them, especially the older ones like Fluke 732A, involves an analog trimmer that when adjusted will start drifting for a while. So what people generally do is just note down the value for every calibration. That way they an apply regression an estimate the drift between calibrations. Now I agree that 60 ppm off nominal is still a lot for a Fluke 732A/B, though I have seen them off by half that. But the standards you make may lack adjustments (adjustments are a source of drift) and you just want to measure the value rather than adjust it to nominal. There is actually very little difference between a 10V and a 9.9V standard for most purposes as long as you know the actual value.

Either way I think that for the fairly high price of the 155 it's a sub-optimal instrument for measuring voltages away from 0V due to its mediocre accuracy of 2% and the analog scale that makes automated recording complicated. I think a 6.5 digit multimeter from a reputable brand will do well enough: as you state the uncertainty contribution will be quite minor, and the 34401A would have 3% uncertainty for measuring 100 uV according to 1y specs. If you want to optimize I'd rather look for a nanovoltmeter like Keithley 181 or later. I do agree that the Keithley 155 is a great instrument for bridges or Kelvin-Varley / Hamon dividers where bias current is a major concern.

[srb1954]

Ok, so that's good news about the null meter. I had a Fluke 845(A/AB?), but I went to my warehouse and it said "find me". I like the idea of using the 6.5s as it seems this will be the most efficient timewise and the cal quote I have for 34401As won't break the bank if I use those 2 and 3456A. I will still need to get another 10V standard to compare to the Guildline.

So far:
Guildline 4410 10V Standard
2nd "reasonable priced" 10V Standard
2x Cal'd 34401A
1x Cal'd 3456A

Does this look like a decent starting point? Guildline Pdf below.

Thanks again everyone.

I have a Startronics VS4 (aka Guildline 4410) and generally find it to be very good although I haven't fully characterised it yet due to problems with my 3458A.

A couple of points to note:
1) The rated ambient temperature range is 16℃ to 28℃. They really mean that upper limit as I find that the oven drops out of temperature regulation at 28.5℃. It still seems to regulate OK below the lower 16℃ limit.
2) The battery life doesn't seem to be quite as good stated in the specs. This will be a problem if you are transporting it to a cal lab unless you enclose it in a well insulated carry case to minimise the power that the oven heaters have to provide.
3) The SLA batteries originally installed are a slightly non-standard size being slightly lower than equivalent SLA batteries available today. The only replacement batteries I could find were 5mm higher that the originals and their terminals are extremely close to shorting to the case. I plan to remove the expanded polystyrene insulation from the base of the battery compartment and thin it down to provide a bit more clearance above the battery terminals.

Thank you for this information. I know you said that your 3458A is not available for testing the standard, but do you have an idea how stable it is with another meter, say a 6.5?

I tested it with a 3456A, 3457A and an 8506A but the results were rather inconclusive as to the actual performance of the Startronics VS4. The 3457A showed a downward drift in the readings whilst the 3456A and the 8506A showed an upward drift in readings. All meters showed significant (5-10ppm) warm-up drift for several months from turn on and also significant (up to 4ppm/℃) variation with ambient temperature in my uncontrolled temperature environment.

All in all, I didn't really get any useful information regarding the ultimate performance of the VS4 from my collection of 6.5digit meters but I hope to have another attempt in a few months. I am currently moving house and have made provision for air conditioning in my new workshop and hopefully will get some meaningful results for testing the VS4. After moving my equipment I will power it on for at least 6 months for any warm-up drift to settle and for any components that might have been disturbed by transport vibration to settle down.

I will also do some more testing with my 3458A to see if behaves better in a temperature controlled environment.

[alm]

I tested it with a 3456A, 3457A and an 8506A but the results were rather inconclusive as to the actual performance of the Startronics VS4. The 3457A showed a downward drift in the readings whilst the 3456A and the 8506A showed an upward drift in readings. All meters showed significant (5-10ppm) warm-up drift for several months from turn on and also significant (up to 4ppm/℃) variation with ambient temperature in my uncontrolled temperature environment.

It would be easier for the meters if you could measure the differences between the four 10V outputs, for example A-B, B-A, B-C, C-B, C-D, D-C, D-A, A-D (so as a ring). This does not exclude the possibility of all outputs drifting the same direction, but it should allow you to get some idea about stability without needing the 3458A. Possibly even better than a 3458A could.

[bdunham7]

I have LM129AH, LT1021BMH-5, LT1027BCN8-5, LT1021CIN8-10, AD588AD, and LTZ1000ACH (saving those). I'm sure I have tons more but I grabbed these for convenience as my warehouse is a mess and I've been trying to organize it for a few years now.

Also, the 1000Hours is a given, as that how datasheets state long term stabilty. I mean in addition to the 42ish days of monitoring. I want to feel confident passed the manufacturers' specs. AN extra 90 days would give the standards 4 months under my monitoring set-up and would give me personal confidence that my standards are solid. Once again, I'm not trying to put anyone out of business, I just want to make some very nice references that will make hobbyist happy.

So you have a variety pack and apparently will be trying a variety of designs.  When I said to start with 1000 hours, what I meant was that after that time your own data will tell you whether the devices have stabilized or not.  After that, your decision as to continue evaluating, reject or ship a particular unit would be based on your own readings. 

If you really are aiming for a 50ppm tolerance, you probably already have enough to start with in your current stable of 6.5-digit meters.  Of course a 10.000000000000000000V golden reference is nice, but you might start by signing up for the USA Cal Club and using the quite remarkable portable reference made by TiN.  Do a 10-day log of your three meters and the ambient temperature and you'll know what you have.  Once you start building some devices and figuring out how to characterize them for tempco, noise, drift and ?, you'll likely find that dialing in those last few ppm is a few steps down the line.

I suspect you'll end up trying to do better than 50ppm since there are competitive hobby-level products already in the marketplace that are well below that.  If you haven't already, check out the 10ppm 10V reference at www.voltagestandard.com. 

 

[denimdragon]

So you have a variety pack and apparently will be trying a variety of designs.  When I said to start with 1000 hours, what I meant was that after that time your own data will tell you whether the devices have stabilized or not.  After that, your decision as to continue evaluating, reject or ship a particular unit would be based on your own readings. 

If you really are aiming for a 50ppm tolerance, you probably already have enough to start with in your current stable of 6.5-digit meters.  Of course a 10.000000000000000000V golden reference is nice, but you might start by signing up for the USA Cal Club and using the quite remarkable portable reference made by TiN.  Do a 10-day log of your three meters and the ambient temperature and you'll know what you have.  Once you start building some devices and figuring out how to characterize them for tempco, noise, drift and ?, you'll likely find that dialing in those last few ppm is a few steps down the line.

I suspect you'll end up trying to do better than 50ppm since there are competitive hobby-level products already in the marketplace that are well below that.  If you haven't already, check out the 10ppm 10V reference at www.voltagestandard.com. 

 

Yes, I am familiar with voltagestandard.com (isn't this Doug's site also? I have one of his 5 volt references).
Thanks for clarifying the 1000hrs. We are on the same page.

I'm putting in my order for LT1013 from mouser tomorrow, but I'm on the fence about the resistors. Let me
give a bit more information about what I have access to.

A few years ago, I bought out a local electronics stores' inventory when they shut down the business. They supplied
most of electronics business in the Texas area so there's tons of NOS components. It's mind blowing. I haven't went through .1% of the stuff there, if that (there's a post on the forum somewhere I believe). The references pictured
came from there. In addition to the tubes, vrefs, op amps, etc., I have a large amout of ww precision resistors, and BMF resistors. Some Vishay, Tellabs, Riedon, Caddock, Micro Ohm, Shall,.. you get the picture. My point is that, I literally have everything I need to make these references at minimal cost, as I've already absorb most of it with my initial purchase of the warehouse.

My goal is to use what I have available, thereby making the references affordable and still be able to recoup some of my investment and make a buck or two. I've been characterising the resistors with Agilent DMM utility (now Benchvue), with my 3456A, 34401A, and 3457A meters. I would like to use these resistors I have for one of TiNs references to see how they perform. However, I would like to buy the recommended resistors from Vishay, but the wait times and price annoys me a bit.

I think I will bite the bullet and buy the recommend parts, then build one with the parts I have charaterized. So that's my current conundrum. What route would you guys try?

I also realized, after sitting in a brown out for the last 5 hours, I'm going to need a nice generator to power my "Mini Metrology Lab" (thanks for that write-up Conrad!) 24/7 without making life long enemies of my neighbors. If that happens when I'm up and running, "the pooch is gonna be screwed"....

[bdunham7]

However, I would like to buy the recommended resistors from Vishay, but the wait times and price annoys me a bit.

Try:

https://webdirect.texascomponents.com/

I'm not sure what I'd try in your case.  I'm trying not to fall down that particular rabbit hole...   Perhaps just build some raw (6.9xxV) reference devices without the divider stage and see how your stability measuring works out. 

[denimdragon]

However, I would like to buy the recommended resistors from Vishay, but the wait times and price annoys me a bit.

Try:

https://webdirect.texascomponents.com/

Wow. I can literally drive there in 15 minutes and it's a staight shot. You made the decision easy. Thanks!

[RandallMcRee]

One thing that occurs to me: it would be value-add if the standards you sold were characterized for noise and, in particular, popcorn noise.

If you are checking for stability this might be part of your QA.

[alm]

I would like to use these resistors I have for one of TiNs references to see how they perform.

TiN is unfortunately no longer active on this forum, but he has been very supportive of such initiatives in the past with design reviews and calibration of self-built standards, so once you have a prototype I'd suggest contacting him.