Prema 6048 teardown

[Kleinstein]

The polarity reversal is via a relay used for current polarity switching. So even if there is thermal EMF at the relay it would not matter. The important parameter for the realy is low leakage when off. Here relays are usually quite good.  A possible weak point with the polarity switching could be leakage current between the supply for the input amplifier part and the ADC+reference part. This could be the transformer and the charge pump part for the ohm mode. A transformer should be resonable well isolated compared to 50 K (20 V range). For the ohm part one could try removing the photomos realys (K13 and / or K14 or the 4 parts to replace them) from the sockets and have a short for the output side (C32) instead. I think this was discussed before, but not sure if actually tested. I don't expect the photomos part to be causing the problems. A quick check of the photomos switches could be by reading a short in the 20 V range and than applying some heat to the photomos switches. If leakage there is an issue heat would normally make it way worse (e.g. > 10 µV of effect for a 10 K rise).

It looks like the zero adjustment is seprate from the gain calibration. The zero adjust point is expected to be done by the normal user from time to time, ideally before each use. I don't think the gain calibration would have much effect of the zero. Nomally the changes in the gain from doing a new gain adjustment are small (e.g. < 100 ppm and often < 10 ppm). The zero point of the ADC itself is shifted to the negative side by some 120 mV. So even of the gain factor is used for this shift it would effect the zero by some 0.1 - 1 µV. 
It would be only a very minor point if one would need to do a new zeroing after doing a gain adjustment. One would have to do it after some time anyway.

The INL problem initially looked like an offset problem, like a mixed up zero constant (for the data from RAX there the missing zero adjustment could be a big part). However the more detailed measurments with more points have a nonlinear curve from about -100 mV to +100 mV. The zero adjust works, but is just in the nonlinear range.  So it is more a problem of the ADC when measuring small voltages  (or possibly low voltage leakage with the photo-mos switches).


I don't see a path that the DMM can directly measure the internal reference. At best there may be a way via ohm current source and shunt, but even that is not for sure. So I don't see a way to have a cal constant for the reference voltage. The cal. constants should be more the ADC scale factor including the reference for the 200 mV / 2 V and 20 V ranges.  This may be just 1 number each and no need for a separate constant for positive and negative sign. The zero / offset should be a separate constant per range and polarity.  Changes are the 6048 has additional corrections based on the temperature measured on the output side. The extra effort for a 2nd ADC only makes sense if it is used for more than just noting the cal temperature. Chances are the temperature dependent part would be a special factory calibration including tests at different temperatures (e.g. 10 C and 30 C ambient).

[Rax]

One has to do a manual zero with an external short at some time after warm-up (can take quite long as there is no fan, but not sure how much the zero drifts with warm up). After that the zero point should be pretty stable. The Zero may even be stable over a limited time, like a few days or weeks. For a new bought meter it would be a good idea to do a careful zero for all the ranges.
[...]
The parts for the pos and neg sign still look linear and with the same slope. Part of the step could still be from an old zero setting or possibly an issue for the KVD (adding some offset) - though not likely.

It looks like the large difference in my case is due to no attempt to zero this prior to the measurements, is that correct? As there are two issues to consider, the intersection with the zero point (large offset, so I need to carefully zero this), and the slope, which seems to be linear and of correct slope?

What would then be a reliable and effective zeroing procedure? Last night I did a quick one with a 3' cable, it was on AUTO and so it went through all VDC ranges. But maybe I need to redo it with the low-emf cable I have connecting the meter to the KVD to collect it readings. I could just short the end of that cable.

I don't have one of those shorting blocks (wide trace between HI and LO and thin trace between SOURCE and SENSE), so I wonder what would be a well thought out zeroing. The manual is not particularly clear on this.

[Kleinstein]

Not having done the zero for a very long time and a possibly bad old value from the prior owner could explain the step in the curve. So the large step may be just the missing zeroing.
After a proper zeroing, repeating the low voltage test in the 20 V range would make sense. A few even low voltages (0 , +-10 mV. +-20 mV) could also make sene in the curve.

For zero one should have V_hi connected to V_lo with a low EMF connection, e.g. simple copper wire. In additon the gurad should also be linked (but low EMF not critical there).
A shorting block as used for 4 wire ohm should be OK too, but is not needed.
As one may have to repeat the zero from time to time it makes sense to get some shorting connector that is reasonable easy to handle without introducing much heat.
One may have to wait for a stable reading before doing the zeroing. For the 20 V range this should still not be too critical. Low EMF gets more of an issue on the 200 mV range.

The short directly at the KVD can also be OK, though not ideal for a more normal use. The test with the KVD assumes that the KVD zero is working - so one could also use that for the test.

[dietert1]

If you calibrate the 20 V range with a 10 V reference with a zero offset of let's say 20 uV going unnoticed, then the slope/gain factor determined by the calibration will be off. As soon as you zero the meter, the 20 uV (2 ppm) will be missing at the upper end.
And if you have a hot relay (an error present for example in the Advantest R6581T) that generates let's say 10 uV thermal EMF, then the + and the - range won't meet ends near zero. As the zero that you create with a short may be in any of the two ranges and you don't know which one, there will be no way to correct that on the host. If the overlap of the + and - range is 120 mV, a zero adjustment using a short won't be reliable.
If the instrument had an indicator whether it is using the + or - range, this could help. Maybe one can mod this. Using a spare digital input (like some meters provide), one could transfer the bit to the host. For voltnuttery you can just ignore the minus range. Don't know if one should disconnect the relay in order to always measure in the plus range.

Regards, Dieter

[Rax]

Here are some readings after redoing the zeroing earlier today. This was with the attached low-emf cables and my kludged shorting block, and included shorting the GUARD. After all the measurements, I see a .00000064V drift of the zero (see pics).

Integration was maintained at 20s to get all the digits.

• 9.99996194V .............. 9.9999712V (this is the closest I've ever seen a reading of this...)
• .1 ratio on KVD ........... .99997780V (2V range, auto)
• .2 ratio on KVD ........... 1.99996355V (2V range, auto)
• .1 ratio on KVD ........... .9999950V (20V range, manual)
• .2 ratio on KVD ........... 1.9999860V (20V range, manual)
• .3 ratio on KVD ........... 2.9999810V (20V range, auto)
• .4 ratio on KVD ........... 3.9999792V (20V range, auto)
• .5 ratio on KVD ........... 4.9999740V (20V range, auto)
• .6 ratio on KVD ........... 5.9999702V (20V range, auto)
• .7 ratio on KVD ........... 6.9999686V (20V range, auto)
• .8 ratio on KVD ........... 7.9999688V (20V range, auto)
• .9 ratio on KVD ........... 8.9999638V (20V range, auto)

• .01 ratio on KVD ........... .0999683V (.2V range, auto)
• .02 ratio on KVD ........... .19999265V (2V range, auto)
• .03 ratio on KVD ........... .29999765V (2V range, auto)
• .04 ratio on KVD ........... .39999582V (2V range, auto)
• .05 ratio on KVD ........... .49999496V (2V range, auto)
• .06 ratio on KVD ........... .59999384V (2V range, auto)
• .07 ratio on KVD ........... .69999080V (2V range, auto)
• .08 ratio on KVD ........... .79998900V (2V range, auto)
• .09 ratio on KVD ........... .89998784V (2V range, auto)

I went ahead and stepped down one more:

• .001 ratio on KVD ........... .01000050V (.2V range, auto)
• .002 ratio on KVD ........... .01999883V (.2V range, auto)
• .003 ratio on KVD ........... .02999932V (.2V range, auto)
• .004 ratio on KVD ........... .03999877V (.2V range, auto)
• .005 ratio on KVD ........... .04999845V (.2V range, auto)
• .006 ratio on KVD ........... .05999756V (.2V range, auto)
• .007 ratio on KVD ........... .06999700V (.2V range, auto)
• .008 ratio on KVD ........... .07999628V (.2V range, auto)
• .009 ratio on KVD ........... .08999589V (.2V range, auto)

[Rax]

I did a quick plot of the low end data for the 20 V range.

How do you go about plotting this? Is there a convenient way (program, website, etc.) to input the readings and have it spit out the graph? I'm dreaming about something like this and there's got to be something out there. I don't have an even "relatively quick" way of doing it.
I really appreciate all the pointers.

[Kleinstein]

Most of the relays in the 6048 are latching types controlled via a series capacitor. So they would not run hot. Normal realys are used a few in the AC part.
Adding a LED to show the polarity (e.g. with low current to the drive side before the capacitor) could help a little.

The 120 mV overlap is the theoretical value.  The actual used overlap is considerably smaller (set by the SW), probably more like 40 mV, as this is the limit of the offset to compensate in the zeroing function. The old INL curve from brandic also has a jump in about that range.
The ADC will not work well for very small values as this involves very short pulses. So one expects a somewhat nonlinear behavior for very low ADC readings. The question is mainly how far the nolinear part extends and if the 120 mV of overlap is enough. The INL test show quite nonlinear behaviour when going slightly negative (up to -40 mV) and thus closer than 120 mV to the limit. The rather linear part only starts at 50-100 mV. So it looks like the overlap is a little to small and should have been more like 200-250 mV.  However the SW did not accept the much larger overlap.


AFAIK the zeroing would do both polarities. It kind of has to as there is no good way to tell the meter which polarity.  At least brandic told that a relay is engaging when doing the zero adjust. So there should be no porblem with having the zero for the wrong polarity. The problem is more a nonlinear response at very low readings (e.g. < 100 mV). So I don't think this is a SW problem but more a limitation for the ADC or less likely leakage from the Ohms part / charge pump.

I did a quick plot of the low end data for the 20 V range.

How do you go about plotting this? Is there a convenient way (program, website, etc.) to input the readings and have it spit out the graph? I'm dreaming about something like this and there's got to be something out there. I don't have an even "relatively quick" way of doing it.
I really appreciate all the pointers.

I used gnuplot. For the last plot the first part as copiing the data to an ASCII file and add the - for the KVD. The inside Guplot the commoand:
fit a*x+b 'C:\Users\fuli\Documents\temp\prema6048 lin.dat' using 1:2   via a,b
plot 'C:\Users\fuli\Documents\temp\prema6048 lin.dat' using  1:($2 - a*$1-b) 
The graph is than coppied to the clipboard and exported as .png.  file from a graphics program.
So the graph is the data minus a fitted line.
Gnuplot may be a bit outdated, but it is free and available for many platforms. (Linux, Windows, VMS)

[Rax]

Wait. Doesn't Excel have this?...  (Graphs based upon tabulated data)

[Kleinstein]

One can also use excel for plotting, at least when the number of data points is small. It gets ugly with 1000s of points.

[Rax]

Here's a first attempt at this.

If anyone had a chance to digest this newer data and has some input, I'd appreciate it. Did zeroing improve things at all?
Scrambling through my day, so no real opportunity for me yet.

[Kleinstein]

With just a few points higher up in the 20 V range there is not much to see. One would need the other polarity to see if the point line up.


The critical range is something like 0 to 200 mV in the 20 V range and maybe 1 or 2 extra points higher up (e.g. 5 V or 10 V) to get the slope.

[Rax]

With just a few points higher up in the 20 V range there is not much to see. One would need the other polarity to see if the point line up.


The critical range is something like 0 to 200 mV in the 20 V range and maybe 1 or 2 extra points higher up (e.g. 5 V or 10 V) to get the slope.

Thank you for unpacking that, Kleinstein. It looks like I'll have to perform those more deliberate steps past handing over the FX to the next custodian, but that's alright. I don't think this empirical research relies on having the FX, so I can continue it based upon my own 731B and other references I have here. Ultimately, one thing I seem to be able to rely on with the 6048 is extremely accurate near 10V voltage readings, which essentially makes the FX optional.

[TUMEMBER]

Something seems odd about the data you provided. Such a rapid change of the error sign in one measurement step is improbable. Maybe it's a bad linearity of the meter or maybe a data error.

[Rax]

Something seems odd about the data you provided. Such a rapid change of the error sign in one measurement step is improbable. Maybe it's a bad linearity of the meter or maybe a data error.

That last step is the direct reading of the FX by the 6048 (not going through the KVD). Not sure exactly how that impacts the plot and drawn conclusions off of it.

In fact, one thing I'm learning about this instrument is how important it is to let it do enough passes and have plenty of integrating time to be as accurate as it can be.

How I know this is that I am extremely close to Illya's FX value by giving it enough time as per above. After multiple passes at 80s integration, it now displays 9.9999678V, which is probably as close as anyone would ever expect this to be to another extremely stable reference that is calibrated to a high standard (the FX, at 9.99996194V). To my math, that's a little over half a ppm of 10V?

Not sure how to interpret all this data but one thing I will do next week is to redo all these readings with plenty of integrating time and multiple passes. I feel that's how the most accurate data will come out.

To clarify, "Output" is the actual reading on the meter. "Stimulus" is what the output from the KVD should be, with an ideal ratio and treating the FX reference value as absolutely true. The last reading bypasses the KVD altogether, and is the direct of the FX on the meter, with the same integration (20s) and only a pass or two at reading it.

[Kleinstein]

When using only a 2 wire connection, the loading from the KVD can make a difference. So the 10  V point directly of the reference can be different from 100% setting at the KVD.

I would not make too much out of getting the same value at 10 V. Everything better than a few ppm would likely be by chance, depending on the last calibration of the 6048. The accuracy of the last adjustment can vary quite a bit, dependen when and where it was done.

[dietert1]

If you think you are measuring the FX to some uV: What happens when you reverse polarity, so that the meter shows a negative 10 V? This is what people do in metrology as a consistency check and to reduce some sources of error. Then the average between the two readings gets used as a result.
Thinking about the 6048 plus and minus ranges: Would be nice if the meter got some means of manual selecting which of the two ranges, i mean similar to "Autorange off" and then selecting a range manually.

Regards, Dieter

[Rax]

New data from yesterday, below. Tempco of the stimulus has got to be much less stellar, being from my DP8200 this time. Temperature has been 24.4C when I started, not entirely sure at the end (forgot to check...).

Two different graphics styles used, hopefully for legibility.

Tried to provide good granularity between -100mV and 100mV, then a couple other points on each side of that at -.9V, -.5V, .5V, and .9V.

[Rax]

Data in .xlsx format is anyone wants a more convenient way to grab it and visualize it better (I am unsure I'm doing that right).

[Kleinstein]

The graph is not really telling much. The meter is still linear enough to high the INL error in the plots.  So see the error one would have to plot something like the reading minus the stimulus or the reading minus a linear fit.
Here is the plot for the output-stimulus.
Looks pretty much similar to the data from brandic.

[Rax]

It looks like the non-linearity at the zero point crossing is about 1.5ppm of reading, then? Which is three times the claimed .5ppm.

If my math is right - and I'm still digesting this - but I don't think I have a huge problem with that. It may not be a $10k competing meter, but it is what it is, and my personal investment in it misses a lot of digits from that other sticker.

It may also be possible it can be modified and improved... We shall see.

(enclosed is my version of the more legible graph generated earlier)

[Kleinstein]

The nonlinearity error is around +-12 µV for the extremes. This is only a little more than 0.5 ppm of the 20 V range.

With the somewhat common an rather similar INL error the question is if this a design fault (maybe only after some design change - the photo-mos switches for the ohm source seem to have changed from 2 x 2 channel one to 4 x channel ones) or a kind of common aging or damage / aging.
The error seems to be relatively easy to find so I would have expected to find such a problem earlier - ideally aready in the design phase, so that they could have fixed it (e.g. with more overlap)
If the error is known and reasonably constant over time and temperature one could apply a numeric correction. No ideal, but maybe some workaround.

[Rax]

I also think my sample is right on the dot at 10V, so I assume its last cal used that stimulus to software-adjust it. From that perspective, I think I need to be watching my expectations with its accuracy as I depart from that value, and what happens at 0V is essentially a worst case scenario. Also, this is in the 20V range, is it expected that does this applies similarly to all ranges? I guess I can map that out later this week.

[Kleinstein]

I would expect a similar behaviour in the other ranges too. For the 200 mV range it may be a bit difficult to see as the zero offset can be relevent in this range.
A test in the 2 V range could really make sense. If the problem is from the ADC side the error should also shift to smaller stimulus values (so +-5 mV instead of +-50 mV range for most of the nonlinear part). If the error would be from the photomos switches the stimulus range should be about the same. So still +-50 mV in the 2 V range.

[Rax]

I am trying to take a step back and gain a bit of a wider perspective on this issue.

I can't really find much on the expected linearity (+,0,-) of the meter. The tightest spec I can find states .5ppm of reading, and/or of maximum reading, (cumulative), which makes the observed magnitude of the error failing at worst just by a hair (does it even, really, fail? I'm not quite sure I can clearly find that in the specs). It also seems like better calibrations (factory, etc.) can make this marginal failure either marginally pass or fail less. In any case, it seems to me all these cases are marginal failures.

Am I misunderstanding this?

[Rax]

I've also checked my meter against David Partridge's test, and get:

• step 4: 10.000069V
• step 6: -10.000050V
• step 8: 0.000092V (!)
• step 11: 9.999974V
• step 13: 9.999973V

So I don't think my meter is inflicted by that issue, is that correct?