Buyers beware, misleading 34470A reference modules (clones) on eBay.

[TheSteve]

Are those pins actually pogo pins or do they just look like it?

[View[+]Finder]

Are those pins actually pogo pins or do they just look like it?

Pogo? No go. The pins are designed to be soldered to wire in a mil spec type cannon connector. There is a shoulder on the pin that, in this implementation, acts as a stop for the ref board. I soldered some cut off pins in place of the wire and fitted these through the eight holes in the MB. Once soldered in place, the pin ends were trimmed leaving the ref board capable of being removed if necessary.

Some explanation of photos posted above:
1) Before, so I would know which way the old ref went
2) Closeup of sacrificial pin being soldered
3) The bottom side of the ref board, the pins go into the holes on the MB
4) The pins sticking out of the underside of the main board, ready for solder
5) The new ref board in place--see the gold pins?
6) The mounting pins with the board removed--see the stops?

Yes, they look massive--not so much so when trying to handle them with fumble-fingers.

[TheSteve]

OK, in the close-up of the bottom of the ref board they look like POGO pins that have had solder added(at the shoulder) so they don't compress anymore.

[View[+]Finder]

OK, in the close-up of the bottom of the ref board they look like POGO pins that have had solder added(at the shoulder) so they don't compress anymore.

Exactly. Minus the spring and the detent. No, just gold pins to give me something to solder in place. Thinking back, I could have used wire, but then what about dissimilar metals and you know someone would call me out on that . . .

[View[+]Finder]

KS34465 with '70 ref board. Latest results from overnight testing . . .


Another test today, 2 hours of data

UPDATE: invalid data, plots removed

[Dr. Frank]

Please indicate, how you have set up this measurement:
-What is the 10V reference?
- Have you connected 34465A and Keithley 6510 in parallel to the 10V reference? (picture of setup?)
- What is the NPLC setting on both?
- how are the data points connected ? (please show a drawn line between them)

There seem to be some sort of oscillation from one value to another.. there's something fishy, either with your 10V ref (not stable) or your setup (ground loop or interference from one DMM to the other)
Frank

[View[+]Finder]

Please indicate, how you have set up this measurement:
-What is the 10V reference?
- Have you connected 34465A and Keithley 6510 in parallel to the 10V reference? (picture of setup?)
- What is the NPLC setting on both?
- how are the data points connected ? (please show a drawn line between them)

There seem to be some sort of oscillation from one value to another.. there's something fishy, either with your 10V ref (not stable) or your setup (ground loop or interference from one DMM to the other)
Frank

At TIN's suggestion I double-checked everything. It looked like there was a problem with the reference (under-volt supply, my fault), however I have just now isolated the source of the divergent data to a mishandling of plotting in the spreadsheet program. I'm working with over 50,000 observation for each data series and using a "line plot with markers" as a starting point with the lines being removed later for clarity. To investigate, I switched the plot type to "2D Scatter" and the divergence disappeared.

That said, the 34456 is showing more drift than in earlier testing and much more than the 6510 on the same reference. The reported temp for the 34465 is 30.7C, the lab is 24.2C, the thermistor in free air 5cm above the case is 32C. I also have an old-style glass lab thermometer stuck inside the 34465 case through the fan opening: it reads 43C with the Hg bulb in the area of the new '70 ref. Temperatures have been stable today.

Graphs and photo of setup attached.

Thanks for your help,
Donal

[TiN]

Oh, a photo of a setup I was so waiting for
So many wrong things on the photo, that hard to pick one to start with.
Probably first - toss the blessed awesome14 "reference" into a bin where it belongs. It was a laughing stock some years back. Don't get me wrong, LT1021 is a nice chip and nothing wrong with it, but it's not adequate to characterize 34465A's performance, even with LM399. I would suggest to build simple LM399A reference (there are many designs available in open hardware) and use that to test 34465A mods. Or if you want something already proven and finished, get one of Ian's PDVS2mini units, which can provide you with adjustable low-noise source up to 10VDC.

Then your wiring need work. Ditch any adapters and bananas where possible. Both DMMs should connect to reference source using own cable, not daisy chain from one meter to another. While usually it works fine at 6.5-digit level, its just bad habit. For most sensitive cases (e.g. unbuffered zener reference) even fancy 8.5 digit meters with AZ may cause issues when used in parallel.

Put reference in some isolated box, that will help to dampen thermal variations a bit. Also since you have another DMM6500 - collect ambient temperature and plot graphs together so it's possible to observe presence/absence of possible correlations.

Your data shows 22 ppms change over 70000 points (hint for future plots - use time scale, not arbitrary samples), which hints that something is very very wrong.

[View[+]Finder]

Oh, a photo of a setup I was so waiting for
 get one of Ian's PDVS2mini units, which can provide you with adjustable low-noise source up to 10VDC.

Then your wiring need work. Ditch any adapters and bananas where possible. Both DMMs should connect to reference source using own cable, not daisy chain from one meter to another. While usually it works fine at 6.5-digit level, its just bad habit. For most sensitive cases (e.g. unbuffered zener reference) even fancy 8.5 digit meters with AZ may cause issues when used in parallel.

Put reference in some isolated box, that will help to dampen thermal variations a bit. Also since you have another DMM6500 - collect ambient temperature and plot graphs together so it's possible to observe presence/absence of possible correlations.

Your data shows 22 ppms change over 70000 points (hint for future plots - use time scale, not arbitrary samples), which hints that something is very very wrong.

Hello Illya,
Thanks for your good advice and also your opinion on bananas (when I get time, I have copper fork terminals and 4-conductor twisted to build proper cables), however, let's not let "perfect" defeat "good enough" just yet. I've been an Ian Johnston fan for years and have his first model as well as his improved version. The mini is a cheaper version of the latter, right? I switched to the Calibratory ones for convenience, I'll try Ian's  . . .

Right now my objective is to determine whether the 34465 with the '70 reference has an error that can be remediated by a proper plastic shield as Dr. Frank suggested or your suggestion that the connection pins might have "tempco" issues. Clearly the 34465 data is out of line with that from the 6510 by an about that might be traceable to "bad wiring"
so it will be easy to test that.

As for the quality of the references, I use the "good, bad, ugly" rule and try to eliminate the "bad" if I can. If you have a suggestion or specification for a reporting format, time scale, etc. that would be helpful. I come from a world of standard deviations, skewness, kurtosis  and R-squared measures to characterize variance in data. PPM seem to hide some easily explained risk in an aggregated statistic.

Thanks again,
Donal

[Andreas]

I come from a world of standard deviations, skewness, kurtosis  and R-squared measures to characterize variance in data. PPM seem to hide some easily explained risk in an aggregated statistic.

Hello Donal,

I think that you will find the Allan Variation as useful tool to describe stability.
There is several Software around which calculates the stability diagrams like
"Plotter" from Ulrich Bangert or "Stable".

Plotter is available e.g. here:
https://www.bartelsos.de/messtechnik/df6jb

with best regards

Andreas

[View[+]Finder]

I come from a world of standard deviations, skewness, kurtosis  and R-squared measures to characterize variance in data. PPM seem to hide some easily explained risk in an aggregated statistic.

Plotter is available e.g. here:
https://www.bartelsos.de/messtechnik/df6jb

That program is exactly what I have hoped to find! Most commercial offerings these days are either bloated (Excel), limited (Numbers from Apple) or slow (Libre Office) and their graphics look like their customer in still in short pants. A quick read of the Plotter documentation got me on the path. I'll download and test ASAP.

Herzlichen dank,
Donal

Sad note: the creator of the Plotter, DF6JB, went silent key in 2014 at age 59. Great program, thanks from KF6OXI

[View[+]Finder]

Here's a run from yesterday afternoon pitting the 34465A with the "Silver" reference (not a favorite of TIN) against the 6510 measuring Ian Johnston's PDVS2. Today, I'm switching the references for another run.

Ian's reference is looking pretty good, I know how diligent he has been in his effort to create a reliable voltage reference for a reasonable price.

[View[+]Finder]

The latest for my KS34465 with the '70 reference from Keysight.
Following Dr. Frank's example, I fabricated an enclosure for the reference board (top and bottom). After a day turned and connected to a 10VDC reference (HP3245A), I created a python script for testing. One hundred readings for taken about 30 seconds apart. Each observation comprised an autocal (*CAL?) command, 20 seconds 'sleep', a query of internal temperature, a query of DC voltage (10V range, auto zero ON, NPLC=10), a record of readings and finally a 5 second sleep. A total of 100 observations over about 50 minutes between 8:40 and 9:20 this morning.

Attached are a plot of measured voltage vs time and a plot of temperature vs time. The plots show a lack of correlation of voltage with time and only a 0.3C range in internal temperature. In short, noise. My next test will involve changes to cables and reduction of EMF sources to see what effect that has on noise.

Some help would be appreciated.
What is the source of the internal temperature measurement in the 34465? What about when the '70 ref board is installed?
I have a thermistor 5cm above the 34465 case, should this be incorporated in testing or is the internal measure sufficient?
I know from TIN and others that LED lights are an EMF source and twisted Kelvin with fork spade copper connectors are preferred, anything else?

Thanks in advance,
Donal

[Kleinstein]

A cycle with ACAL will add some extra noise, as this involves extra convertsion to measure the scale factor. I don't know how long ACAL actually takes, but one may have to extect soemthing like 0.1 ppm "Noise" for an ACAL call, as this would be still better than the nominal resolution.

To check the reference dirft, one should have a run without ACAL calls an with a stable temperature. If at all one would call ACAL in longer intervalls, like every 2-6 hours. Ideally one would than have enough data before and after to calculate back the jump from ACAL.
Chances are one would want to full resolution / minimal noise. So there should be more like contineous readings for some 100 PLC (maybe extra averaging of shorter readings).

The HP3245 is sopposed to use a LM399 ref. So don't expect anything very low noise. It may still be a selcted (e.g. through out the really bad ones) one. This ref would still be not really good enough to fully judge the reference drift or noise.

[maxwell3e10]

Regarding enclosure to prevent air currents, I had some good luck just using tape over the analog section of the meter
https://www.eevblog.com/forum/metrology/turning-fan-off-to-reduce-dmm-voltage-noise/msg2652360/#msg2652360

[View[+]Finder]

A cycle with ACAL will add some extra noise, as this involves extra convertsion to measure the scale factor. I don't know how long ACAL actually takes, but one may have to extect soemthing like 0.1 ppm "Noise" for an ACAL call, as this would be still better than the nominal resolution.

Autocal on my 34465 takes 10 seconds and another 5 seconds for the "OK" message to clear. The Python script sleeps for 25 seconds, however I can't tell whether this includes the autocal time. In any case, the DMM is quiet for at least 10 seconds before a measurement is taken.

At your suggestion, I will try interspersing autocal every 10 measurements and see if that makes a difference.

[Kleinstein]

ACAL after only 10 measurements (still depends on the integration time) would be still very fast and likely a little to short to really see the steps.

ACAL likely does quite a few things (different ranges, maybe even some AC) so there would no be not so much time for the gain calibration for the 10 V range.
One may want a little waiting ACAl for settling - even though there seem to be no such problem as with te 34470. The 5 seconds waiting for the Ok to clear may aready be the settling time for this.

To get most data and thus least noise I would skip the sleep part, but take more data. If looking for possible popcorn noise of the references, it may help to no average to much and accept a longer file. 10 PLC data are probably a good compromise to look for noise from a LM399 or simialr ref. . One can allway avearge later, but there is no way back to finer time resolution.

[View[+]Finder]

Just to confirm, ACAL is the program that I run on my HP 3458a as suggested before starting a set of measurements. It runs for quite a while and tweaks everything to the internal temperature. The 34465 has an AUTOCAL command that takes only seconds to complete. I don’t know exactly what it does, however I assume it is intended to compensate for temperature changes as well.

Last evening I started another run with autocal every 100 readings. It is at about 8,000 of 10,000 observations now and should complete in a few hours. Based on the data so far, there has been a decline in temperature (nighttime normal) and a rise in voltage with the change in voltage 90% explained by temperature.

Post results later . . .

[View[+]Finder]

The data capture (7,766 observations) started at 19:30 last evening and is described in an earlier post. Two graphs are attached. The 12.26.02 shows the internal temperature of the 34465 (X) against measured voltage (Y). The temperature decreased from 30.5C to 25.0C (approx) over the night--the (X) values are in the same order as time.

The second graph,12.24.35, shows voltage (Y) over time (X) as represented by the observation number. The sawtooth pattern is an artifact of the operation of autocal every 100 observations. In effect, the steady decrease in temperature resulted in a recalibration and a concomitant increase in the reported voltage measurement.

I'll leave it to those who know about metrology to help us understand what's going on.

[Kleinstein]

The first curve shows a relatively linear temperature dependence. It looks like a TC of some -0.6 ppm/K . This is little high for just a LTZ reference, but not that extreme. Part of the problem could be how the neg. side of the ref is sensed - not sure if the extra signal is detected. Chances the same pin is used for the normal ADC converstions and only Autocal may look at the different pin.

The curve over time shows quite some effect of the ACAL call.   Here it does not look like ACAL correctig temperature drift, but more like some settling after the ACAL call - so a little like the ACAL problem of the 34470 !

[View[+]Finder]

Just finished reading all of the '70 ACAL error posts--some really good work was done by forum members on this issue! Dr. Frank's post points out the need for a full range DCV calibration to get the ratios corrected. Looks like that is my next task.

Thanks for helping,
Donal

[Dr. Frank]

Just finished reading all of the '70 ACAL error posts--some really good work was done by forum members on this issue! Dr. Frank's post points out the need for a full range DCV calibration to get the ratios corrected. Looks like that is my next task.

Thanks for helping,
Donal

Exactly. It was also my main goal, to search for the root cause of this ACAL failure by inserting the LTZ reference into the 465A.

Your continuous ACAL engagement during your measurements also does not seem to provoke any shift errors.

For targeted stability analysis, i.e. over temperature or time, you should better do appropriate test runs w/o ACAL.

I also estimated a too high T.C. of about 0.5ppm/°C for the LTZ board, this will be directly tested on the  board, outside the 465A.
Maybe you could also do the same with your original reference board.

Frank

[View[+]Finder]

Just a general FYI

When extracting data from Keithley 6510 and 6500, I use the web interface over ethernet and the measurements are 13 digits for voltage. With the Keysight 34465, I use a USB stick and get 9 digits for voltage--the same when using SCPI in a Python script. By comparison, my 3458 provides 10 digits in a Python script.

Questions:
What happened to the data dump software for the KS344xx DMM's? I used it a lot in years past and now can't find it. (ditto, my glasses)
Has any one had a look at the hex code for the firmware to see what differences might be in place for the several meters?
How many digits of actual information are in the SYST:TEMP reading on the 34465? To me it looks like everything past ##.# is suspect.
Is it worth the trouble of putting a thermistor (the little silver cylinder, not a chip) in the 34465 to get a reading?
If the 34465 main board is the same as the one in the 34470, how do we know that the voltage ref connections are ok for the '70 ref mod in the 34465?

Whew . .

[TiN]

eBay 34470A module temperature coefficient benchmarks

We often hear from gray bearded metrologists and analog designers’ importance of using stable and good components in analog circuits, especially voltage references. Hundreds of hours were spent in debates about voodoo PCB slots, swastika-shaped LTZ fanouts, vacuum-insulated thermal chambers, cadmium-enriched solders and quadruple shielded PTFE wiring. But often such discussions lack of any quantitative performance data to actually show benefits of all that effort. Here we address this important issue in terms of thermal stability measurements of voltage reference module shown above.

Keep in mind, reference module tested here is NOT manufactured by Keysight Technologies. This is aftermarket clone copy of original design, with subpar assembly quality and unknown source of the parts. That said, I do not have 34470A original module to compare 1:1, but based on my experience with LTZ circuits I do expect similar results.

Equipment used to perform experiment is ubiquitous characterized Keysight 3458A 8½-digit voltmeter, DIY temperature chamber with 40W Peltier heater/cooler, Keithley 2510 TEC controller, calibrated Fluke 1529 Chub-E4 with Omega RTDCAP-100A-2-P098-050-T-40 sensor and our own xDevs.com teckit Python application to control all equipment and process data into DSV-file for further analysis.

Temperature in chamber was swept gradually from +18 °C to +50 °C with speed around 0.03 °C per minute with static step at 50% of sweep (+34 °C) for 1 hours and 1 hours for peak temperature. Ambient room temperature and humidity stability was maintained at +23 °C ±1°C and 15% RH ±2% respectively.

Reference was powered with Keithley 2230-30-1 linear power supply with supply voltages +12 and -12 VDC. Current consumption from module was ~26-29 mA on positive rail and about 3mA on negative rail, which is typical and expected for LTZ1000A datasheet circuit. Module was assembled with LTZ1000A chip manufactured in 2018 week 38 with legs cut flush to the PCB surface. None of this voodoo LTZ standing on long kovar legs rubbish . Resulting plot reveals output voltage variation with change of temperature inside the airbath box.



Reference performance was quite bad +0.114 ppm/°C (coefficients α = 0.107 ppm/°C and β = -0.0003 ppm/°C^2) with onboard resistors as is shipped from eBay, as I already expected. This is about 3 times worse than good LTZ reference should have.

Good LTZ1000A design capable to provide thermal stability better than 0.05 ppm/°C and this is what I have used as a target during testing and tweaking of all my xDevs.com KX, xDevs.com FX 10V and xDevs.com QVR low-noise DC Voltage references. With some effort stability better than 0.03 ppm/°C can be achieved from typical datasheet circuit, without slots, cadmium solder or trimming LTZ1000A kovar legs trimming to micrometer precision. After all, LTZ1000A is power in/power out device and require just careful system design.

Typical 1% SMT chip resistors as used by seller have temperature coefficient from 100 to 250 ppm/°C, not the 2-5 ppm/°C expected by LTZ1000A reference. Now idea is to replace cheap resistors with something much better and repeat same temperature stability test to narrow down if the excessive tempco caused by resistors or something else like poor PCB component placement, SMT-type LT1013 or routing of the copper traces around.

I bought some reasonably-priced Susumu SMT RG and URG series resistors in 1206 size, specified to have tempco better than 2 ppm/°C.

Digikey part number	Susumu part number	Resistor description	Price
408-1788-1-ND	URG3216L-101-L-T05	RES SMD 100 OHM 0.01% 1/4W 1206	$8.38 USD
408-1645-1-ND	RG3216L-102-L-T05	RES SMD 1K OHM 0.01% 1/8W 1206	$5.20 USD
408-1648-1-ND	RG3216L-153-L-T05	RES SMD 15K OHM 0.01% 1/8W 1206	$5.20 USD

Of course, there are even fancier metal foil VPG resistors such as VFCP Y1630 (over $11 USD/pcs) or FRSM Y4023 (over $18 USD/pcs) but in standard LTZ1000A circuit I doubt we will see much difference in DC Voltage output temperature stability compared to more affordable Susumu 2ppm/°C parts. Long-term stability is another aspect, but its completely out of scope of this article. In the end of the day Keysight 34470A is a benchtop general-purpose DMM and it is not suitable or supposed to act as a laboratory DC Voltage reference equipment.



Pay attention to twice smaller scale on vertical ppm axis. Benchmarks show significant 257% improvement after simple resistors swap. Now reference performance matches expected LTZ1000A circuit capability, with box tempco just +0.046 ppm/°C and slope parameters α = 0.0431 ppm/°C and β = -0.0005 ppm/°C^2

Same data presented in RAW timescale format also presented below, together with some resistor tempco runs. I often run multiple experiments and measurements at same time and combined plots showing multiple unrelated values, which my friends and people can find often confusing. Our DUT +7V reference measurement presented in olive color chart line on this example plot. This is a data from a temperature run with Susumu resistors and presented to show the timing of the experiment. Temperature used for tempco calculations is shown in maroon color (from Fluke 1529 thermometer).



P.S.To provide justice for the seller, I have contacted seller about misleading resistors discrepancy and after discussing performance drawbacks of the unit's resistors as in eBay listing (with photos at time of purchase, seller replaced photos with cheap resistors already) seller issued full refund on my two modules. Buyers beware, as with everything that comes from secondary market!

[View[+]Finder]

Attached are plots of supplemental testing of the autocal sawtooth in my 34465 with a Keysight '70 reference. The top plot was captured between 9:00 and 10:30, the middle one between 11:00 and 12:30 and the lower one between 12:30 and 14:00. As suggested, the last one was done without intermittent autocal. Autocal was run at the start of the sequence with a sleep of 25 seconds prior to measurements. The voltage source was the PDVS2 on battery power from 16AHr lithium 4S.