Prema 6048 teardown

[Rax]

The FX value, for anyone looking to see the deviation of this Prema from it.

[Rax]

A question for branadic, or maybe others - does it look like the calibration procedure for this meter saves independently constants for each range and mode?

I'm considering doing a quick calibration with the FX I have at hand in the 20V range, and use my KVD to divide that 10V down for the 2V and 200mV calibrations. I could possibly use my DP8200 for other ranges and DCA (or just leave alone the old cal for those ranges), but only provided recalibrating the ranges above would not discard the constants for the ranges I'd (intend to) leave undisturbed.

[Kleinstein]

The calibration should have constants for each range, maybe only a divider ratio for the 200 V and 1000 V range. Due to the relay way of changing the polarity, the gain should not change. The offset seems to be adjusted once before use. So proper warm up before the zro adjust can be important.

The 20 V range seens to show some 5 ppm discrepancy with the FX reference. Not clear how much of this is due to the reference and maybe settling of meter internal reference after a longer time of no using it.
It depends on the quality of the last adjustment though. Chances are that the meter still has the factory cal data and no later adjustments done.
Chances are the 2 V range could be about as accurate. So I doubt a calibration with a KVD or to the DP8200 would improve things. An adjustment would likely only make things worse. Chances are the 6048 is still way more accurate than the DP8200 even though the calibration is 9.5 years old and the last adjustment even longer.  So unless there is a damage / lost cal data or one has a really good reference one should not do an adjustment.  Even many cal labs could adjust to the full factory specs.

The KVD and DP8200 would only be a check. I would more consider it a check of the DP8200 and KVD if they give plausible votlages.

[Rax]

So I doubt a calibration with a KVD or to the DP8200 would improve things.

Thank you, Kleinstein. My plan is not to adjust this with the KVD + DP8200 (at least not what my question is about, apologies if I wasn't clear).
The plan is:

• Adjust the 20V range with the FX reference directly
• Adjust the 2V range with the FX divided down with the KVD. For instance, a .1000000 ratio from the KVD would output .999996194V out and into the DMM which is at roughly 50% of the range, so well within a recommended stimulus for setting that range
• Adjust the 200mV range with the FX divided down with the KVD. For instance, a .0100000 ratio from the KVD would output .0999996194V out and into the DMM which is at roughly 50% of the range, so well within a recommended stimulus for setting that range

I'd probably not touch any other range. At least not at this time. My statement on the DP8200 was just to mention I could do other ranges with the DP8200 (no KVD involved, at least I don't see the benefit there), but maybe later if at all.

The huge benefit with the steps above would be standard transfer from the FX. I don't think I'd find another opportunity for an 8.5 meter adjustment/calibration (that's also affordable - but this far all my nearby cal labs have declined adjustment for this, and most have actually declined cal altogether).

Even after the adjustments above, I could still restore the factory cal data from the other memory with the procedure in the manual. The only thing I have now which I risk losing with the steps above is the calibration from 2003 (which, you're right, may have just been plain cal, no adjustment involved).

[bdunham7]

I'm considering doing a quick calibration with the FX I have at hand in the 20V range, and use my KVD to divide that 10V down for the 2V and 200mV calibrations.

I would not change the calibration constants at this point.  The meter appears to be within the 1-year spec @10V.  You haven't checked it for mid-term stability, it has just been shipped and may have been dormant for some time, and also you don't have it in a temperature controlled environment.  How much of that 4-5ppm apparent deviation is due to tempco is unknown at this point--on one hand they claim a 'typical' tempco of 0.05ppm with the advertised compensation scheme, but the specs list a maximum tempco(outside the +/-5C window) of 1.0ppm. 

Edit:  that was the 6047 spec...

[branadic]

The 6048 has factory cal constants stored in its EPROM, that can be restored at any time with the unit turned off, setting the switch at the back and turning the unit on. This writes the factory cal constants from the EPROM into the NVRAM.
The current calibration constants are stored in the NVRAM. If adjustmentshave been made in the past you will find them here.

So if you consider to adjust some of the ranges I suggest to backup the NVRAM with the As Is state first, so you can restore them at any time into the NVRAM.

Second, restore the factory cal constants and again backup the NVRAM, so you now can compare both and see if adjustment have been made in the past.

Third, if you then adjust some of the ranges, e.g. based on the cal constants from the first backup written back into the NVRAM, you can afterwards make another backup of the NVRAM content and intercompare the three to find the constants that have changed.

I also suggest to check the age of the NVRAM, it's maybe worth replacing it with a new one or with FRAM such as FM16W08.

-branadic-

[Kleinstein]

Unless there is another change to the hardware, I don't see a need to do an adjustment. It is not sure that the FX reference is more accurate than the meter. For the KVD this may also not be so clear.  It could still be well worth it to check the 2 V range against the KVD and FX reference. The DP8200 may be OK to check for the linearity in the low range - not all units may have the same low voltage problem. The INL error observed by brandic looks so obvious that they should have noticed (and ideally fixed - e.g. with a little more shift to neg. side) in developement.

The 6048 stability is sensitive to the offset drift of the OP77 and LT1007 in the ADC. So when exchanging the OP77, one would have to readjust. The offset of the OP-amps is effecting the ref. voltage seen by the ADC and would thus be relative to 7 V. This should have the same effect on the scale factor of all ranges (V, amps, ohm).
A backup of the NVRAM and maybe an exchange is indeed a good idea.

[branadic]

Let's discuss based on facts, not on assumptions or expectations.

1. There is a bug in the firmware, that is, with fixed voltage range zeroing is not working as one would obviously expect, the unit is not changing polarity. That only happens if the meter is set to auto range mode.
2. Three different units show identical bad INL behaviour. Anyone can come up with a good explaination why that is.
3. During adjustment of the 20 V range I can observe on all units that if I adjust with 5% of range and check at 50% of range, still being in adjustment mode, the 10 V is already way off.

What we make out of that?

-branadic-

[bdunham7]

2. Three different units show identical bad INL behaviour. Anyone can come up with a good explaination why that is.
3. During adjustment of the 20 V range I can observe on all units that if I adjust with 5% of range and check at 50% of range, still being in adjustment mode, the 10 V is already way off.

What we make out of that?

I'd make out that it's not good, but can you quantify those errors?  For bootstrapping backyard-Bob type calibration such as I do, being able to transfer a single reference up and down by relying on things like the meter accurately reading 1V on a 10V range is key.  If that doesn't work, I'm out of luck.  So how much discrepancy is there between a 1V and 10V reading on the 6048?

Edit:  Ok, so I read back through the thread and I see your data.  I'd just say that this meter joins the Fluke 8500A series in that INL makes up a good portion of the error tolerances.  I couldn't find any express statement of guaranteed INL in Prema's specs or manual.  So yeah, 10:1 or 1:10 transfers are not going to be good.

I'd suggest Rax set up his KVD with the FX reference and his KVD and see what the readings are for ratios 0.1, 0.2 and so on up to 1.0. 

[Rax]

I'd suggest Rax set up his KVD with the FX reference and his KVD and see what the readings are for ratios 0.1, 0.2 and so on up to 1.0.

I can do that, but I do have an ask - I hope we'll be taking this as an additional piece of information on the Prema 6048 saga. None of the instruments I have here and would use for this are infallible (just like research relies on peer review) - so perfect linearity would be a great sign, but imperfect linearity may be due to any one of the components involved. All useful/additional information though.

[bdunham7]

I can do that, but I do have an ask - I hope we'll be taking this as an additional piece of information on the Prema 6048 saga. None of the instruments I have here and would use for this are infallible (just like research relies on peer review) - so perfect linearity would be a great sign, but imperfect linearity may be due to any one of the components involved. All useful/additional information though.

Yes, either very good results or large enough errors will answer some questions, but intermediate size (small but not small enough to ignore) errors are inconclusive and just raise more questions.  But that's what we're here for, right? 

[Kleinstein]

The shown INL curves very much point to a problem of at low voltages, like below some 100 mV in the 20 V range. As a rough description this leads to all readings higher than 100 mV to be some 10 µV shifted up. So for 1 V in the 20 V range this would be some 10 ppm, for a 10 V ref. only an 1 ppm effect.  The INL curve would let us expect that the adjustment at 1 V would be about 9 ppm off (to low a reading at 10 V).  What was the amout of error observed ? 

The INL problem at the low voltages is just at the very edge of the ADCs range. This is a range where the reference is active only for very short times ( e.g. 0.15 % PWM ratio for a 120 mV shift of the zero point). I don't know the modulation frequency, but it should not be much below 1 kHz and thus a pulse length of some 1.5 µs or less. This is in a range where the OP77 at the intergrator may not yet be fully settled.  Residual voltage at the integrator input can lead to an INL error. Something like 1 mV for 1 µs would be enough to explain a 10 µV error.

There may be a way to make the DMM accept the larger zero shift: the manual gives a limit of 0.2% of FS and thus 40 mV for the offset correction in the 20 V range.   A shift in the hardware of a little less than 40 mV (e.g. ~ 150 K or 220 K parallel to R36) has a good chance to be accepted by the firmware. From the INL curve the error would be already down to about half from this smaller shift.
There is a small chance that this 40 mV limit is relative to the last setting. So one may be able to make it accept a larger shift in steps. One could test this by a simpe repeated Zero with something like 30 mV , 60 mV , 90 mV  at the input. If this works, chances are it could also be used to make it accept a larger shift in the hardware.

[branadic]

The shown INL curves very much point to a problem of at low voltages, like below some 100 mV in the 20 V range. As a rough description this leads to all readings higher than 100 mV to be some 10 µV shifted up. So for 1 V in the 20 V range this would be some 10 ppm, for a 10 V ref. only an 1 ppm effect.  The INL curve would let us expect that the adjustment at 1 V would be about 9 ppm off (to low a reading at 10 V).  What was the amout of error observed ?

The answer to that is already given in the data package I attached to this post:

https://www.eevblog.com/forum/metrology/prema-6048-teardown/msg4652476/#msg4652476

Attached a quick difference to S7081 plot of that files for 20 V range with factory constants, 20 V range adjusted at 1 V (5% of range), 20 V range adjusted at 10 V (50% of range) and 20 V range with factory constants again.

-branadic-

[Kleinstein]

The difference between the 1 V cal and 10 V cal curve is around 12 ppm.  So that is still just in the range that could be explained by the INL error seen at low voltage. So I would not consider this a much different problem.

[Rax]

Attached a quick difference to S7081 plot of that files for 20 V range with factory constants, 20 V range adjusted at 1 V (5% of range), 20 V range adjusted at 10 V (50% of range) and 20 V range with factory constants again.

-branadic-

Can you please detail both the exact measurement and the units on both axes of the graph? Secondly, I assume the "InitCal1" and "InitCal2" are factory cal constants during two subsequent restorations? Thank you very much in advance.

[Rax]

Here are some measurements with the FX and an ESI RV722 KVD.

I used an integration time of 80 seconds - which made it very tedious, but I feel it'd return the most reliable results - and have not attempted to zero anything (which I don't think this meter is very peculiar with). Connections from KVD to P6048 are low-emf and shielded and separately guarded (AB-Precision cables), those from the FX to the KVD are just regular banana (all copper mating surfaces). Ambient temperature has been around 23.5C.

A verification short with one of these banana copper cables returned a tiny fluctuation around .0000000V (last digit fluctuating up to maybe 5 on both sides of zero).

• 9.99996194V .............. 10.0000395V (a little further delta from expected than we've seen before... fyi, it's been running since I got it)
• .1 ratio on KVD ........... .99998300V (2V range, auto)
• .2 ratio on KVD ........... 1.99996792V (2V range, auto)
• .1 ratio on KVD ........... 1.0000690V (20V range, manual)
• .2 ratio on KVD ........... 2.0000616V (20V range, manual)
• .3 ratio on KVD ........... 3.0000561V (20V range, auto)
• .4 ratio on KVD ........... 4.0000533V (20V range, auto)
• .5 ratio on KVD ........... 5.0000479V (20V range, auto)
• .6 ratio on KVD ........... 6.0000449V (20V range, auto)
• .7 ratio on KVD ........... 7.0000412V (20V range, auto)
• .8 ratio on KVD ........... 8.0000414V (20V range, auto)
• .9 ratio on KVD ........... 9.0000365V (20V range, auto)

Looking forward to the surgical analysis....   

[bdunham7]

It looks at first glance like yours has the exact issue Branadic describes, if I'm understanding things.  Also your 2V and 20V scales don't match, but I think that is secondary for now.  Could you do another set of readings on the 20V range using ratios of 0.01 to 0.10, and the same with reversed polarity?  You could probably shorten up that integration time quite a bit since there's no point in wringing out every last digit at this point.

[Rax]

Could you do another set of readings on the 20V range using ratios of 0.01 to 0.10, and the same with reversed polarity?  You could probably shorten up that integration time quite a bit since there's no point in wringing out every last digit at this point.

4s integration on the 20V range (manual) at 26C, positive readings:

• .01 ratio on KVD ............. .100098V
• .02 ratio on KVD ............. .200099V
• .03 ratio on KVD ............. .300098V
• .04 ratio on KVD ............. .400098V
• .05 ratio on KVD ............. .500097V
• .06 ratio on KVD ............. .600097V
• .07 ratio on KVD ............. .700096V
• .08 ratio on KVD ............. .800096V
• .09 ratio on KVD ............. .900095V

Negative (leads reversed):

• .01 ratio on KVD ............. -.100088V
• .02 ratio on KVD ............. -.200090V
• .03 ratio on KVD ............. -.300088V
• .04 ratio on KVD ............. -.400087V
• .05 ratio on KVD ............. -.500087V
• .06 ratio on KVD ............. -.600086V
• .07 ratio on KVD ............. -.700086V
• .08 ratio on KVD ............. -.800085V
• .09 ratio on KVD ............. -.900084V

[bdunham7]

As you can see you have a ~.00009V constant offset that reverses itself with polarity.  Just like the 3 units Branadic measured. 

What does it measure with a shorted input?  If this is a design issue, I don't see how it could ever meet its specifications of 5ppm + 1ppm/range (20uV).   

[Rax]

Just for fun, I did the .2V range too. 4sec integration, auto-ranging.
positive readings:

• .01 ratio on KVD ............. .09999710V
• .02 ratio on KVD ............. .19999321V
• .03 ratio on KVD ............. .3000058V (jumped range)
• .04 ratio on KVD ............. .4000043V
• .05 ratio on KVD ............. .5000032V
• .06 ratio on KVD ............. .6000018V
• .07 ratio on KVD ............. .7000000V
• .08 ratio on KVD ............. .7999989V
• .09 ratio on KVD ............. .8999969V

Negative (leads reversed):

• .01 ratio on KVD ............. -.09999920V
• .02 ratio on KVD ............. -.19999524V
• .03 ratio on KVD ............. -.3000072V
• .04 ratio on KVD ............. -.4000056V
• .05 ratio on KVD ............. -.5000040V
• .06 ratio on KVD ............. -.6000022V
• .07 ratio on KVD ............. -.7000009V
• .08 ratio on KVD ............. -.7999998V
• .09 ratio on KVD ............. -.8999978V

[Rax]

If this is a design issue, I don't see how it could ever meet its specifications of 5ppm + 1ppm/range (20uV).

As far as I know, these things were supported until recently, at least in Germany (branadic may have more info on this). My own unit says "MFR SPECS" (what that means exactly, is anyone's guess... ).

The obviously question is who's willing to rewrite the firmware in order to fix it. Second question - how do we adjust these afterwards to pass calibration.

[Rax]

What does it measure with a shorted input?

Earlier today: "A verification short with one of these banana copper cables returned a tiny fluctuation around .0000000V (last digit fluctuating up to maybe 5 on both sides of zero)." Though I'm not quite seeing this right now (seems to fluctuate between hundreds and tens of nV at different times of the day?). I wonder if all this is due to some stability issues (leaking, etc.). 

[dietert1]

With other DVMs we have Autorange and Autozero. For calibration one usually turns off Autorange. Calibration needs to be done per range. Sometimes one calibrates one range and keeps previous divider ratios, e.g. in AutoCal.
With this meter there is a similar problem with zero calibration and/or Autozero. Calibration needs to be separate for the + or in the - range. To calibrate the +20V range one would for example adjust at +100 mV and at 10 V and the - 20 V range at -100 mV and -10 V. So there should be four calibration constants. Looks like they have one constant (slope) instead of four.
Maybe one can implement this calibration on the host. Does the meter tell whether a measurement was taken in + or in - range? Just the sign of the result? Does the meter implement separate Autozero for the + and - range?

Regards, Dieter

[Kleinstein]

The meter seems to no support a classical auto zero. 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 zero should be separate for both polarities (offset from the ADC and offset from the input amplifier) and AFAIK the meter does both togehter - there would be essentially no good way to tell it which polarity.
As the polarity reversal is with a relay one may not need a separate calibration for the positive and negative sign. The point around zero is adjusted with the zero adjust and not just calibration. It could still make sense to have both - more like a test, that usually passes.

Getting quite difference readings in the 2 V and 20 V range is a bit surprising. The zero adjustment does not look perfect (difference between the polarities), but the zero error still looks small

As you can see you have a ~.00009V constant offset that reverses itself with polarity.  Just like the 3 units Branadic measured. 

What does it measure with a shorted input?  If this is a design issue, I don't see how it could ever meet its specifications of 5ppm + 1ppm/range (20uV).   

The error that brandic saw is on the order of some 25 µV  extra 'step' in the +-50 mV region. So this would still be just in the +-1 ppm of the range.
I would consider this rather high to be a design issue, though having nearly the same type of error with 3 meters is suspecious. There could still be a variation of the parts used (e.g. different batch of OP77).

The data from RAX look much worse. I did a quick plot of the low end data for the 20 V range. The attached graph shows about 10 x the step. 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.

[dietert1]

If the + and - range is by a good relay with low thermal EMF deviations, then this explains the good symmetry between the + and - ranges.
Some way or the other the zero adjustment gets lost during calibration. Apparently their calibration formulas don't preserve the previous zero adjustment but some other voltage, maybe the previous reference voltage.
Does the instrument have a calibration constant for the internal reference voltage? If you protect the internal reference voltage as a calibration point and the reference has drifted due to aging, the slope adjustment will destroy the zero.

Regards, Dieter