Prema 6048 teardown

[Andreas]

Andreas, someone populated 100R instead of 1meg, must have picked up the wrong resistor I guess???

Mhm,

not shure if it is a mistake.
in the early 1986 databook the resistor wasnt there at all.
https://www.eevblog.com/forum/metrology/vintage-ltz1000-from-1986-nib-!/msg1827785/#msg1827785

I think the 100nF * 10K (input resistor) = 1 ms time constant is enough for a heater.
The 1 Meg resistor adds some measurable noise visible with the naked eye/scope at the heater pin.
At least I do not want a "noise generator" near my reference. So I have added additionally 10nF between output and input of OP.

My guess is that someone "improved" (really?) the cirquit at the time when the A-version was introduced.

with best regards

Andreas

[dietert1]

In general a thermostat without proportional term is near unstable due to the delay between heater and sensor. I am writing delay because it isn't a time constant, but a higher order low pass. Probably somebody tried how large the proportional term can be made for the LTZ1000. And the observed noise may not be noise but real: the thermostat working against air drafts.
The LTZ1000 heater is very fast. Without any bandwidth limiting in the regulator i once saw it oscillate at about 50 Hz, so the delay time from heater to sensor is about 10 msec.

Regards, Dieter

[Andreas]

The LTZ1000 heater is very fast.

Hello,

I think you mean the LTZ1000A?

So the best way to check this would be either to stimulate the Setpoint by a small amount and see the reaction.
Alternatively examining the power up / power down sequence like in my LTZ1000 measurements e.g. here:

https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg832030/#msg832030

If there is any instability on the heater output which is not related to the setpoint you should see it on power up/down.

with best regards

Andreas

[Kleinstein]

I can not see the heater delay from the trun on transients shown in the link (have not checked the whole long thread  though). It seems the supply is coming up too slow to get an instant temperature reading.

One can estimate the delay for the heater from the dimensions: from the die picture the distance from the heater (outer ring) to the sensor is about 1 mm. The thermal diffusivity for silicon is at 87 mm²/s  - so expect a little over 10 ms as a dead time before power from the heater to reaches the sensor.

The PI adjustment rules (e.g.  Chien, Hrones und Reswick) want the time constant in the PI regulator to be something like 2.3 - 4 times the delay for good reaction to disturbance.
From my experiance shorter can lead to instability, while longer slows down the response a little. The thermal layout external to the chip would not effect the delay very much, It can effect the best gain factor for the regulator. Even there is mainly changes the heat loss and not so much the heat capacity, which is the main factor effecting the permissible loop gain.

So one should expect a suitable time constant somewhere in the 50 ms range. Shorter may be problematic, a little longer is more on the safe side.
The LTZ1000 standard circuit calls for 100 nF and 1 M and thus 100ms , which sounds like a good, safe start and about right.

[branadic]

Replacing the 100R by 1meg (and the LT1078 by LT1013) didn't change anything. Next step is to wait for the resistors to arrive and replace the 120R first, repeat the measurement and second place the ratio divider.
For the measurement in the thermal chamber the board was and is wrapped into some non-conductive white foam, so airdrafts are already prevented.

-branadic-

[Andreas]

Replacing the 100R by 1meg (and the LT1078 by LT1013) didn't change anything.

I did not really expect any large change by this.
OK the LT1078 is a bit slower and has a little less open loop amplification and current output but should be good enough.

I would check the 1/f noise next.
Perhaps we have a "little jumper" (popcorn noise) or a too low zener current because the 120R is not conductive. (you could measure it across the diode).

with best regards

Andreas

[essele]

Drifting slightly off the current discussion, just wanted to report that my Prema 6048 has just suffered a SRAM failure (Error 8 ), I had just ordered a replacement M48Z12-70PC1 when I realised it was the same pinout as the CAL ram in the 3458A.

So I can now report that it appears to all work nicely with an FRAM adapter and an FM16W08 FRAM (which is partially annoying as it's too late to cancel the other order!)

I do need to run through a home cal again as I've obviously lost the prior cal (again!), and I will confirm that everything is properly stored, but I'm certainly not getting the Error 8 anymore.

I not sure how to read the date code from my M48Z12. Branadic's one from earlier in this forum has the form H99XXYYZZ, where YYZZ is year/week. Mine is missing that, but that does suggest that W1728X is NOT a date code, therefore my next best guess is 1999.

Details of the adapters I used are here, with the zip file of KiCad files a couple of posts down...

https://www.eevblog.com/forum/metrology/3458a-upgrade-questions/msg2951062/#msg2951062

Cheers,

Lee.

[branadic]

I had mine replaced right after the unit arrived. Battery was still okay, so that I was able to read the content. However, good thing is, that you don't loose factory calibration.

Today I received my resistor kit, thanks to the help of some board members here. Resistors have now been replaced, reference is warming up and will then be monitored and measured over temperature.

-branadic-

[meandeev]

Hello,

can somebody take some detailed pictures around the power-supply at the preamplifier board? I have a "jumping" 6048 on the bench, which starts 1..5Volts too high when cold and after 2hours warm-up it is within 50-60ppm to a reference, but also jumping 10ppm up and down.

So I started as usual with the power-supply. The +27V rail is real +32V and I saw a serial connection of a zener (with 30V!) and a normal diode for CR6 (on the schematics it is 27V). But it seems genuine, the soldering is like other original one.

I think, that saw at the pictures of branadic also something like that, but the resolution is too bad.

[branadic]

So you bought that 6048 recently off ebay?
The construction is indeed original, no need to worry, though not documented in the manual.

-branadic-

[meandeev]

Hello Branadic,

yes I bought it. Meters with this precision come not up in ebay-europe very often. It was not a bargain (especially with these errors), but I couldn´t  resist and it should be fixable.
The +32v on the +27v-rail made me nervous.  But if you say its ok, then I´m fine.

Thanks!

[branadic]

That was the second 6048 within a few month on ebay:
https://www.ebay.de/sch/i.html?_from=R40&_trksid=p2334524.m570.l1313&_nkw=prema+6048&_sacat=0&LH_TitleDesc=0&_fsrp=1&_odkw=prima+6048&_osacat=0&LH_Complete=1&rt=nc&LH_Sold=1

Give it the normal electrolytic replacement and check for solder joint cracks, which was the fault on my unit - at the bridge rectifier, that caused some intermittent restart.
Can you please report if you have a separate reference board or the LTZ already installed on the ADC board? I'm asking as my unit is serial number 1027 and yours is 1035, so they are pretty close and I wonder at what point they had the new board revision installed.

-branadic-

[meandeev]

I have ser-no 1085 and it has the "new" (?) PCB with the LTZ onboard. One can find IC´s dated from 8706 to 9011.

I saw the other auction too, but it was too risky to me because of the description: "does something" - whatever that means.
This one seems a litte bit better because of the photo of the resistance-measurement - so it must have passed the selftest.
But there was definitely somebody in - look at the not cleaned soldering and some white residue and the lifted pad on the integrator PCB . It was repaired by Prema in 2015 - but I doubt that this bad soldering was done by them.

[branadic]

Interesting, it looked like 1035 on the back.
Someone really messed inside, that is not looking genuine. The white residue propably is IPA. Good luck on repair attempt.

-branadic-

[meandeev]

one more question to the 6048 owners:

which AC voltage do you measure at the input of the rectifier CR1? Iset the device to 240V mains and measure 38.4V at CR1. This comes down to 21.7V DC over C1 and is the input for U1 (7815 voltage regulator). I ask, because this regulator is hot after half an hour (ca. 60-70°C, I cannot touch it with my fingers for a long time).
So is this normal? I think I have to open a separate repair thread...

Thanks!

[branadic]

It seems that the feets for Prema multimeters are available at Farnell, at least they look identical:

https://de.farnell.com/okw-enclosures/a9257209/klappfuss-unterbau-schwarz-pk4/dp/775484
https://de.farnell.com/metcase/a9257207/klappfuss-unterbau-grau-weiss/dp/775472

Just in case someone is in the need for them.

-branadic-

[branadic]

Hi folks,

I'm in the process of characterizing (INL), calibrating + adjusting a Prema 6048, but am somewhat stuck at the moment.
The Prema 6048 is somewhat special, as it detects polarity at its input and sets the input relays accordingly, but also switches to a third position if the input is zero.
During calibration only zero and only one input value (something between 5% and 100%, preferably between 50% and 100%) is asked for to adjust a range. There seems to be no gain correction, which is kind of weird and leads to a large INL error, obviously similar for positive and negative polarity, but with a totally different value for zero. To reflect that issue I performed a quick 1 V step INL measurement with the factory constants.
Has someone experience with adusting a Prema 6048 and could give me a hint?

-branadic-

[Kleinstein]

The error shown does not look like a gain error, but more like an offset problem. The slope is the same for the positive and negative sign. Somehow the zero seems to be OK in the middle - I don't see how the 3rd mode for near zero readings is done. It looks like there is a tiny bit of offset and both modes should be able to read zero, though not much room for hum and the like.

The polarity reversal with the switches also looks like it should not cause any change in the gain. The same gain constant for both signs makes sense. There is however a possibility for offset errors, for both the input part (amplifier) and the ADC part.
Some 12 µV is not that much for the ADC and could come up over the years.
Chances are a zero offset correction (for the part behind the reversal, so it could be the ADC itself) step in the calibration could correct much of the shown INL error.

[branadic]

The meter was shorted/ zero'd before taking this measurement and it was powered up for quite a few weeks too. So no, that's not the answer to the observation.
I already adjusted the meter at 10 V (50% of range) and confirmed the same INL error afterwards.
I then adjusted the meter again, first by adjusting the 10 V range at 1 V (5% of range), then checked at 10 V, adjusted there and verified the 1 V point again, were I could the observe a shift to the prior set value for 1 V. Whatever I do, this behavior is reproducable.
I would totally understand if 5% and 50% to 100% were used to adjust one range, the INL curve would then be flat, but that is not how the meter behaves. Hence why I ask here for someone with experience. It seems like something is missing in the manual.

-branadic-

[Kleinstein]

The descritption of the calibration is rather short and split over 2 parts (7.11 and 9) and there may be something missing.  In the description in chapter 9 I am missing a zero adjustment. Under 7.11 it calls for zero steps. These may be needed, even if the meter already ready zero.

I would expect to have a seperate zero constant for the ADC and than one for each range for the input amplifier.  It is not that clear when the extra zero for the ADC is done - this would need something like a zero reading with both polarities, and ideally switch the polarity relay during a zero phase. It may be just 1 ADC zero for all ranges - a partial calibration may miss this.
Some drift for the ADC would not be that surprising. So I would expect this to be part of the normal CAL procedure or maybe even a part repeated with a normal Zero command without the switch in CAL mode.

[branadic]

Within the last days I have tried several different things, like zeroring as part of the adjustment as well as zeroing priopr to adjustment. However, the INL curve is always the same. So the manual is missing something important.
An email to Prema was yet without any answer. I will try to give them a call next, otherwhise I'm running out of ideas. 

-branadic-

[Mickle T.]

Prema Instruments has used its patented unipolar ADC type in almost all of its products. Perhaps the user manuals for other models have more detailed information about zero calibration? For example, Prema 6030/6040.

[branadic]

Thanks Mickle T.
I had a phone call with Prema and their DAKKS cal lab today, so we are currently in the information exchange process. Question, where did you get this 0.1 ppm INL number from?



-branadic-

[Mickle T.]

...Question, where did you get this 0.1 ppm INL number from?

This is a very old table and contains several inaccuracies. But as far as I remember, I took the value of non-linearity from the datasheet on the PRI5610 ADC and publications from the bbs.38hot community. I understand that this is not official data, but I did not have a better estimate at that time.

Is the non-linearity problem described above similar to the one already discussed?
https://www.eevblog.com/forum/metrology/possible-problem-with-prema-6047-or-6048-meters/

[alm]

Prema Instruments has used its patented unipolar ADC type in almost all of its products. Perhaps the user manuals for other models have more detailed information about zero calibration? For example, Prema 6030/6040.

I checked the Prema P6031A manual (an older, late-eighties 6.5 digit DMM), and it's pretty similar to the P6048 manual. It sounds like zero calibration is updated every time you zero the meter, and not part of factory calibration. They don't specify a zero calibration as part of the adjustment. I guess they relay on auto-zero to null any offsets, and consider linearity inherent in the ADC design. I didn't find anything about linearity in their theory of operations, except that how their lack of switching produces better linearity compared to other designs that switch during integration (cf. HPAK Multislope I/II/III).

The P6031A manual is a bit more detailed in its description of the polarity detection, in that it mentions the turnover is within 1 LSD. But of course 1 LSD at 6.5 digits is a bit more lenient than at 8.5 digits .