Prema 6048 teardown

[Kleinstein]

The very low frequency noise data are not very relaiable, but from the curves shown in the datasheets the OPA140 is a bit higher in the 1/f noise than the OP07 / OP77 and definitely higher ( ~ 5 x) in the voltage noise than the LT1007. The OPA140 may still be good enough.

If the OPA140 is hard to get, one could in most cases use an OPA141. The main difference should be more stringent testing and a usually better offset (may also be just testing but could also a more accurate trim).

[branadic]

Could you add all of your changes into the schematic? It seems I couldn't follow all of them completely. I would then try to recreate them and provide confirmation.

-branadic-

[Rax]

Could you add all of your changes into the schematic? It seems I couldn't follow all of them completely. I would then try to recreate them and provide confirmation.

-branadic-

dietert1: +1.

[dietert1]

The nonlinearity test with -160 mV to 160 mV sweeps in the 20 V range also looks clean. Vertical scale of the diagram is +/- 1 ppm of 20 V range. The MAX5719 test DAC has an INL specification of 3 ppm, so that would be 0.48 uV in 160 mV.

Yes i will prepare a step by step document of the two mods that made a difference.

I also have an image of a DIY scanner (work in progress). A friendly person supplied SDS latching relays - similar to the other relays in the P6048. Instead of the Prema C32 microcircuit there is a CD4094 shift register with latch and the CD4049 buffers were replaced by complementary emitter followers. There is a supply supervisor made from discrete parts and a CD4011 to produce the inhibit signals for the two decoders.

Regards, Dieter

[Rax]

The nonlinearity test with -160 mV to 160 mV sweeps in the 20 V range also looks clean. Vertical scale of the diagram is +/- 1 ppm of 20 V range. The MAX5719 test DAC has an INL specification of 3 ppm, so that would be 0.48 uV in 160 mV.

Yes i will prepare a step by step document of the two mods that made a difference.

I also have an image of a DIY scanner (work in progress). A friendly person supplied SDS latching relays - similar to the other relays in the P6048. Instead of the Prema C32 microcircuit there is a CD4094 shift register with latch and the CD4049 buffers were replaced by complementary emitter followers. There is a supply supervisor made from discrete parts and a CD4011 to produce the inhibit signals for the two decoders.

Regards, Dieter

Dieter, this looks very promising, and I'm quite a bit blown away by your results. Fantastic work!
I'll seek to run these trials with my unit so we'd confirm the results in at least three different samples.

[branadic]

What is now left open is a full sweep to see overall INL. However, the result looks very promissing. 

-branadic-

[dietert1]

I know you can do that with your calibrator and your other meters.
It's team work anyway.
Let's mention perdrix (David Partridge) who discovered the nonlinearity in the first place.
And Kleinsteins comments have been important. Concerning OPA140 noise, probably the zener noise dominates in that part. I just used what i had.

Regards, Dieter

[Kleinstein]

I had a few thought on the noise. The op-amp U10 is only relevant for the time the reference is connected. U9 (the intgrator) is also mainly important when the reference is active. Both OP-amps work togehter with the reference signal that does to the input of U10.  So one kind of get a combination of the noise of the 2 OP-amps and the LTZ1000 reference. In this set the reference is usually the larger noise source by quite a bit.

Another point is that the time the 3 are connected depends on the input signal / voltage. With the variable time this gives a noise that gets higher the higher the voltage, not just from the principle low frequency reference noise, but also from higher frequencies ( > 99 Hz). The crude picture gives an added noise part proportional to the square root of the voltage.
With the relatively slow 20 s integration the extpected contribution is small - more like hard to see. If really an issue the point would be more adding a little filtering to the reference. The expected noise constribution is still small and though simple it is likely not really worth it.

The 2 main expected noise sources are jitter from the PLL - so the modification with the VCO range may help somewhat, though probably not much as there are also internal limitations.
The other point could be quantization, depending on what filtering is used in the custom prema chip. For the 2 V range the input amplifier can also contribute quite a bit, though it is still not that bad.

I have not seen detailed noise data on the 6048, but from the few curves shown at the start of the thread the noise is not that bad and with a 10 V test signal not that much higher than the 2 references involved.

[Rax]

I know you can do that with your calibrator and your other meters.
Regards, Dieter

Dieter,
Please use the "Quote" feature of the forum. I personally don't know who you mean to respond to there.

It is teamwork, and if it all pans out, this may improve on the OEM performance of an 8.5 digit meter (read that a couple of times, and let it sink in). I don't want to jump the gun - and let's definitely do the work - but if this works, it'd be pretty groundbreaking. I'll congratulate everyone contributing when we get the peer review completed.

[dietert1]

Here i made a quick writeup, let's see whether it serves.

Regards, Dieter

Edit: One image slipped away..

[dietert1]

Here is a schematic including mods of 50 Hz => 3.6864 MHz PLL.

Regards, Dieter

[branadic]

Thanks. In the end only the opamp replacement made the difference, is that correct?

-branadic-

[dietert1]

You can try that. If it isn't sufficient, next i would try 4 to 6 . Then 12 and 13, then 9 and 10. I documented the other mods as they are visible in the images and they are meaningful. Probably i won't go back to start to find the minimal solution.
For example two independent sources (Lee Essen and an expert at Systems Engineering in Stolberg) talked about voltage regulator failures. The instrument works without fan! Also remember Kleinsteins report of thermal oscillations. What i did certainly looks unusual. An alternative solution would be a common heatsink mounted to the frame that serves to shield the mains transformer at the same time. But mounting all seven regulators on a common heatsink will introduce capacitive coupling and isolation voltage limits between the different power supply sections.

Regards, Dieter

[dietert1]

Here i did a noise test with the modded instrument versus our 14.1 V 5x2 LM399 array. Vertical scale is 14 uV = 1 ppm of reading. The 20 seconds reading noise is 0.2 uV (= 0.01 ppm of 20 V scale) yet there is some additional process at the time scale of 3 or 4 minutes that makes the total standard deviation 0.87 uV. Probably i will try protective covers for the LTZ1000 reference inside the Prema.

Regards, Dieter

[Kleinstein]

The noise does not looks like normal white of random walk type noise. It is more like quite some popcorn noise included.  Some of the jumps look a bt large to come from 5 x LM399 in parallel, as this should give jumps of some 0.8 - 1 µV.

A cover over the LTZ1000 would definitely make sense, though I doubt it would solve the issue with the jumps.

[dietert1]

One can see slopes with 15 or more readings in a row, so that would be a process of 5 minutes.
Maybe the variations we see are caused by 50 Hz mains frequency steering. One could try to run the instrument with the PLL off. Or log mains frequency for comparison. Some meters include monitoring mains frequency.

Regards, Dieter

[branadic]

I've replaced U9 and U10 with OPA189 in the white Prema 6048 (no other modification), as this is what I had at hand. While I'm enjoying the sun, INL measurement is running, so we will know more tonight. By the way, U10 was an LT1007 (plastic DIP package) in that unit, the OP77 was a nice CERDIP.

-branadic-

[Kleinstein]

I don't see a mechanism of the mains frequency or a drift rate in the frequency to have a direct effect on the reading. An unstable mains may add some extra noise, but I don't see a reason for steps.
My main suspect for the steps are the references. Mybe there is something to couple the LM399 references to each other and this way get steps of 2 x the expected size.

[branadic]

INL measurement finished, we can clearly see the massive improvement. The remaining INL error visible might be due to the INL of the S7081 itself, which Frank measured to be ±0.3 ppm in its 10 V range or the effect of temprature changes during the measurement.

-branadic-

[Rax]

INL measurement finished, we can clearly see the massive improvement. The remaining INL error visible might be due to the INL of the S7081 itself, which Frank measured to be ±0.3 ppm in its 10 V range or the effect of temprature changes during the measurement.

-branadic-

Good stuff, branadic. I think your only mods are U9 and U10 to obtain these improvements, correct? I'm currently looking into what suitable op amps I can source and fit with a DIP adapter to replace those two ICs, but it's going to be a few weeks given I'll be mostly away from the bench due to conferences and work-related duties across the US. 

Question for both branadic and dietert1: have you attempted to assess how these mods influence the state of calibration of the unit? Does it stay within the same margins as before the mod, or, should these changes be implemented, the unit would definitely need to be sent in for calibration/adjustment? I wonder if in the pursuit of linearity at zero-crossing, one would effectively discard the original/ulterior calibration constants' usability.

[Kleinstein]

A change of the OP amps will effect the calibration. The offset of the OP-amps add to the effective 7 V reference voltage. This is not only the offset of the new OP-amps, but the difference to the old ones.
This would be one factor effecting all ranges (including resistance). So there would be a chance to check the scale factor before (e.g. 10 V referene voltage , 10 K resistor) and than at least get an idea on how much change happend.

[dietert1]

My unit definitely lost calibration when i installed the PWM filter (mod 9) effectively incrementing the 5 KOhm precision resistor by 120 ohm. It's a good idea to transfer the P6048 calibration to other standards before applying mods.

For me the next step is working on the LTZ1000 reference. The Prema 6048 runs it's reference with a 12.4/1 KOhm divider for the oven setting. I will sweep its set temperature and adjust for zero TC. The test DAC and the software application i made for the nonlinearity sweeps can be used with little change. I will just pull R80 402 KOhm and use it to sweep the set point. TC compensation will also affect voltage calibration.
And it will also affect the total TC of the instrument. The real thing will be reengineering the firmware to understand how temperature dependent calibration works.

Regards, Dieter

[Kleinstein]

Unless the TC of the reference is really high, I would not change anything with the reference circuit unless there is a real problem.  A 12.4 K and 1 K divider gives a relatively low set temperature and this can cause a relatively low maximum room temperature.  If it works and is well settled a low set temperature also has advantages. The old age of the reference is a good thing here.

The real thing will be reengineering the firmware to understand how temperature dependent calibration works.

This is another reason of no changing the reference TC - the current firmware should compensate for much of it.

[branadic]

I already replaced the cheaper oven resistor divider by a metal can resistor network in the past, together with a metal can resistor for the 120 Ω, but haven't aimed for zero t.c. back then.

The real thing will be reengineering the firmware to understand how temperature dependent calibration works.

That would indeed be the real deal to finally fix temperature effects. I guess they used some common temperature compensation constants on all their units, rather than individual temperature compensation, which is a bit odd.

I will replace the opamps on my black unit by OPA189 today and perform the INL sweep, but I'm sure we will see the very same improvement. Once done I guess it is worth repeating the INL measurement for both units against our 3485A under controlled and stable lab temps or maybe even in a climate chamber.

-branadic-

[dietert1]

The noise test log i had above continued for another 24 hours. Looking at a temperature sensor inside the test DAC, i found a temperature variation of -1 °C during the night caused about 1.5 uV of drift of the readings. So the difference between the TCs of the Prema 6048 and the LM399 array is roughly -0.1 ppm/K. To separate the TCs, one of the devices should move into the temperature chamber.

Here i have a first temperature sweep result of the LTZ1000 reference in our instrument - as is. Resistor R80 402 KOhm was not present. For the sweep i used a resistor 403K from the test DAC (-4.5 to 4.5 V). This gives a sweep range of about +/- 5 K. The default voltage of 0.5 V at the base of the temperature sense transistor causes a slight asymmetry of the sweep.
The plot looks as expected with a negative TC of the voltage readings of the LM399 array, which means a positive TC of the LTZ1000 reference of about 31 ppm/K. The curvature of the line fit residuals is also as expected. Adjustment for zero TC will require a resistor of about 13 Ohm in series with the zener.

Regards, Dieter