Advantest R6581 8.5 digit DMM mini teardown/repair

[Kleinstein]

The Datron 1271 circuit is a little odd, as the time constant is still rather low. The slower one is only 10 ms and thus way faster than the real DA and the integration. So it would only effect the fast part, that is already suppressed by the use of more slopes and thus some delay.

To get a suitable time constant of longer than a 10 PLC conversion (e.g. 1s) it would need a really large resistor, so more like 100 M and 10 nF. This would be more similar to the Datron 1281, where 18 M and 47 nF are used.  I don't think it would need a second time constant - more like a temperature dependent strength. If the strength of the compensation is too large one should see the effect in the INL, especially the more or less sharp step close to zero, when the rundown pattern changes.

[Mickle T.]

Playing with a time constant gives some ambiguous results. It was possible to reduce the "+" INL (from 0.8 to 0.3 ppm), but at the expense of increasing "-" INL from 0.1 to 0.7 ppm 
All additions dismantled.

P.S.
I found a some oddity in the slope amplifier: it have a very small input resistance (R219+R220) and easily can overload the integrator op-amp U206. Even a large 4.7 kOhm resistor, connected to I.OUT, can overload the U206 and makes it nonlinear.
So I suggests another DA-comp scheme.

[szszjdb]

Hi, Mr. Mickle T.,
Thanks a lot!

Have you got any progress with the DA compensation modification?

My unit is still the same with the 0.5ppm error of INL at positive range after burn in for 10 days. It seems nothing could do now for me.

Best Regards,
szszjdb

[Mickle T.]

I did not continue experiments and completely disassembled the DMM.
After a zillion attempts to fix the unfixable bug with +1/+10 V and 1/10 MOhm transfers, I gave up.

[Kleinstein]

The shape of the INL curve does not really match the expected error due to DA. So the observed INL error is likely not due to DA - maybe a small contribution, but not that much. So I would not have much hope in DA compensation to give much of an improvement.

From all the experiments so far, most where kind of excluding some effects. For most of the part we can exclude:
1) DA  (observed INL curve does not follow average integrator voltage)
2) capacitive coupling to Q204/Q205 drain (observed INL does not follow frequency for comparator pulses)
3) rundown related effects  (the INL error is not much higher in 1 PLC mode)
4) Input switching related effects (e.g. settling of the amplifier) (the INL error is not much higher in 1 PLC mode)
5) input current to U205 (OP177 at input of integrator) (not much change when going to an OPA140 instead)

There was a slight indication that loading of U206 did have an effect - though it did make the error larger. Still it is not very clear how this small extra  load should have an effect.  My best guess here would be something like shifting the ground level at some points could be a factor here. In addition to a static ground level shift there might be small (e.g. 10s of µV) AC contributions involved - so it may not be so easy to see those. For finding the INL souce I would suggest something like a careful INL test with an without extra loading to U206 (e.g. 5 K from I.out to ground or -15 V) and maybe with a larger value for R220 (this would reduce the current flow between U206 and U207 at the cost of slightly higher noise).

Another part that may be worth testing is the error with large resistors. One point to check here is to test a few more resistors (e.g. 100K, 200 K, 500 K, 700 K) instead of just 1 MOhms in the 1 M and 10 M range.  Another possible check here would be to see if the current source is really linear - this would need a second meter to measure the test current in the 1 M / 10 M range with and without a series resistor and thus with a different voltage at the source output. If leakage around the current source is causing this the error should be even larger in the 100 M range. So if a suitable instrument is available one could check those currents.

[Mickle T.]

Another part that may be worth testing is the error with large resistors. One point to check here is to test a few more resistors (e.g. 100K, 200 K, 500 K, 700 K) instead of just 1 MOhms in the 1 M and 10 M range. 

I was already did it with all of the 1, 10, ... 1E6 Ohm range. Current source is extremely linear and don't have a measurable INL. Also I was check a currents settling problem with a 10 nF capacitor, added to the 900 k current shunt resistor (R500). After ACAL I was got the same 200 ppm shift. Changes of the U405 compensation string R*C405 also gives nothing. Decreasing by half of bootstrap resistors R406 gives nothing. Aforemention U107 mod gives nothing. Disabling of the D402 (overcurrent protection) also gives nothing, as well as variation of ambient temperature in large range.
I tend to believe that the problem is more firmware, than hardware. For example, due to a different Xilinx XC3042A configuration, loaded from XC1736E serial EPROM. XC3042A does in real time all low-level A/D signal processing, DMA and many other things. Why not?

[szszjdb]

Hi,Mr.Kleinstein and Mr. Mickle T.,
Thanks a lot!

I found that some of the INL is coming from my voltage source for testing ,the DIY LTZ 7-10v source. The former might be not so stable  and I reassembly a new one yestoday. The other contribution might be the 0.022uf cap parallel on the DMM when using with the resistor string, as I had took them to filter some noise. And also for the single voltage capturing, I using 20 point avarage instead of just 1 estimation. And also some modification before like adding the decoupling cap on each OPs, it likely helps to the final result. The update INL curve looks much better than before, which I had never imaged. But there still have 0.3-0.4ppm error in the middle of  positive range and it is more like the performance of others 6581.
I will try to add some load to U206. I am also considering about the leakage of the MUX when in high positive voltage input , but can not prove it.  This might lead to the INL error over the 0.1ppm limit, right? I suppose that the fresh MUX have few leakage and lead to the ideal INL curve as the datasheet. When suffered by the surge in some condition and the protection circuit malfunctioned at that moment, the leakage of the switch became extreme large and contributed to the INL of positive range  finally.

Attached FYI.

Best Regards,
szszjdb

[Kleinstein]

The INL curve looks not that bad now. Over a large part it is following the curve for the 3458. So if we take the 3458 as a reference point (and not the divider chain) the difference in the INL curve would be the INL. Here values are more like in the 0.1 ppm range over a large part (except for +8 V and higher).  With the resistor chain, the input impedance chances with the taps and this might explain parts of the curve seen - deviation is largest in the center, where the impedance is highest. It might also make a difference if the curves are measured with both meters in parallel of just one meter at a time.

I am not that much convinced the input current is due to damage from over-voltage. If could be some kind of aging shifting the JFETs threshold. Contamination could also be an issue. Some of the leakage could also be due to variations in the FETs used - making some meters better than other from the start. At least some of the input current due to not having a kind of precharge phase and the peak due to the forward biased gate below about -6 V are more like design faults - this is more like the expected behavior for the given circuit, with only small variations between units. Especially the switching related input current is likely not due to a damage, more like a problem by design.

It looks like calculating the effect of input current should be relatively easy. This is true for the quasi DC part for the non AZ mode.  However this is not that case in the AZ mode with current in the form of short pulses on AZ switching. So the effect of input current may be even be a simple linear function of input impedance and input capacitance can also have an influence. So the 22 nF at the input can make a big difference - especially for the comparison the the 1 PLC and 10 PLC mode.  It might be worth to check on how much effect the source impedance has. So maybe use a 7 V source and measure with a few different series resistors (e.g. 10 K, 50K, 100 K, 150 K, 200 K).

From the 604 INL curves it looks like mainly the small range of +8 V and higher is still a problem. This is not really good, but one might be able to avoid this range for most critical cases.

[szszjdb]

Hi,Mr.Kleinstein ,
Thanks a lot!

By adding the 5K from U206 PIN6 to PIN4 , it is likely no more change observed . Attached FYI.

Best Regards,
szszjdb

[Kleinstein]

No change with an extra load is kind of a good sign: at least likely not much INL due to the OPs output stage and also no effect of loading the -15 V.

Having a rather similar curve for the measured INL for the 3458 and the 6581T kind of makes me suspicious that there might be common source in the test source, like the change in output impedance with voltage. So at least part of the observed curve could be from the input current and source impedance and not the ADC or amplifier. So it might be about improving on the test first.  At least the quasi DC part of the input current (non AZ mode) could be measured (e.g. capacitor discharge) and corrected if the source impedance is known. The input current due to switching can be higher an average, but much of this might not have an effect as there is some delay before the measurement starts, so much of that current might be ignored. With the extra 22 nF at the input and a higher source impedance settling can take more than 1 ms and thus more of the switching current would effect the reading.

[szszjdb]

Hi,Mr.Kleinstein ,
Thanks a lot!

The LTZ 7v will change with the different series resistor measured by 3458,attatched FYI.   The cap parallel on input do harm to the INL test as your advised.

Best Regards,
szszjdb

[Kleinstein]

The effect of a series resistor on the 3458 is not that interesting, the 6581 would be the meter to check. The data give some 6 µV drop with 100 K and thus some 60 pA of input current - thus about the normal range.  No need to check much below 10 K, as there are 8.8 K inside the meter. The interesting part would be the range (and maybe a factor 2-3 higher) of impedance that the resistor chain has, thus maybe 10 K -100 K.

However there was an alternative test to see the effect of input currents due to switching: this was the comparison of 1 PLC and 100 PLC mode. The old test was not very high resolution (the 1 PLC mode had limited resolution despite of averaging, normally the data should be good for one more digit). Still that test showed that there was not that much effect to to more frequent switching. If that old test was done with the same resistor chain and no extra cap - this old test would be good enough. The effect not caught in the 1 PLC / 100 PLC comparison is the non AZ mode bias current - this would be relatively easy to measure (there should  be old data on this already) and the effect due to the input impedance can be calculated straight forward.

A slightly odd point is why the apparent INL of the 3458 is that high.

One point possibly worth checking would be the INL in the low voltage range, where the pattern and voltage during the run-up changes (e.g. 0-100 mV range). For the small range the calibrator with a good quality divider (e.g. 1 :50 ... 1:100, could be already inside the calibrator) could be good enough.  Where the feedback pattern changes we expect some step like error in the 0.02 - 0.1 ppm (of the 10 V range) range due to DA.

[szszjdb]

Hi,Mr.Kleinstein ,
Thanks a lot!

I had changed the resistor chain to total 10k to minimal the charge effect. Attached the reading with different series Rin ,FYI.

Re test the turnover error in 1 and 100plc mode, found significant different reading with them.

Best Regards,
szszjdb

[Kleinstein]

The series resistor test (essentially only the 0 / 10 K / 100 K numbers really matter) shows quite some input current. 9 µV over 100 K is an effective current of 90 pA. Not really bad, but also not really good.  The odd thing is that the voltage reading goes up with 100 K in series - so the effect is opposite to the average current measured with the capacitor discharge !

The comparison of 1 PLC to 100 PLC mode is similar to the old data. There is quite some difference, but still not that much to have significant effect on the INL in the 100 PLC mode.  way to look at the data is not the turn over error, but the difference from 1 PLC to 100 PLC mode at the individual voltages. The 1 PLC mode amplifies the effects of the rundown part and the mux switching effects (e.g. input current spikes) by a factor of 10. So 1/9 of the difference would be there contribution to the INL in 10 PLC or 100 PLC mode.  So there is a few µV of difference, but divided by 9 this is not really much.

[szszjdb]

Hi,Mr.Kleinstein ,
Many Thanks !

I re-checked the voltage with different series Rin and it do get larger reading when with 100k. It might be the changing positive bias current flow to the series Rin outside the meter with different  vin.

It seems the MUX have the most contribution to the system INL(0.4ppm MAX) as others had been eliminated.  Although  not perfect as the 3458, it is still can take some work.  So I stop expecting more from my 6581 now.

For my used 3458 , it might have some issue as it tend to have more noise reading when with the 10K resistor chain than 6581. And it will get unstable reading when measure the dirrect -7v of LTZ zener output, likely some oscilation occured.  Also with the a little more turnover error as 2uv. Others looks good with the zero noise and the stable reading when measuring the LTZ +7V . The bias is also very good. The 3458 is too complex to modify and I have no idea to fix something.

Best Regards,
szszjdb

[Kleinstein]

For the input current at the 6581 mux, there is a possible chance to get some improvement, by adding a kind of pre-charge phase. It sounds a little crazy at first, but I think it would be possible. The point would be to add a kind of additional input to the MUX that is activated before the DCV input. The suitable signal it there from the input protection buffer.  There usually is enough time from switching to DCV till the measurement starts. So an monoflop chip could be used to create this extra step in an analog way.  The pre-charge phase would also prevent the current pulse seen due to the slow amplifier.

The early test with the capacitor discharging was showing the voltage to more towards a voltage in the 2 V range (I don't remember the sign). So at 7 V the DC input current sign should be so to reduce the voltage. So chances are hight the higher voltage with the series resistor is due to something like current spikes / setting from switching - still the average current is like an positive input resistance, but there is overshoot. So it is possible the with 100 K series resistance settling is a little to slow and coming from the positive side.

The 3458 input would be an different thread. It would likely start with measuring the input current to see if there is a problem. Even if no intend to fix / change something, it would be a good idea to know / check the input bias also for the negative input.

[szszjdb]

Hi,Mr.Kleinstein ,
Many Thanks !

I will try the adding pre-charge circuit later after work out the detail connection.

For my 3458, I had checked the bias and it is very good arround 20G Rin and found nothing abnormal from the curve. Attached FYI.


Best Regards,
szszjdb

[eurofox]

I just have a question with respect to calibration since my knowledge of the Japanese language reading is not very good 

Of course a basic calibration is done with External 10.0000000VDC and 10.0000000K ohm.

When temperature change do you guy's use the Internal calibration?

It always scares me a little been afraid of changing the basic calibration. 

Internal temperature is always 33° C at least each time I check it.

[Kleinstein]

The internal calibration is there to reduce temperature an aging effects. So it is supposed to be done when the temperature changes significant (e.g. more than 1-2 K) and from time to time, like once a day, if critical measurements are done. The internal cal should be done after the meter is warmed up of cause.

[eurofox]

Is a GPIB driver available for NI or Excel or any other software?

Since there is a GPIB interface I suppose since this instrument was quite popular there should be a way to communicate with it.

I know it is possible to write a srcipt 

[serg-el]

http://www.adcmt.com/techinfo/product/catalog_ducument/pdf/CommandMatrix_6581_1.pdf

Hello. I do not have 6581, only 6871.
you can use this document.

[TiN]

I've started week long run on my resistance references, and this time added my (broken/bad) R6581T into the mix.
Main standards: SR104 on 3458A-rusty, G9330 zombified with VHP resistor on 3458A-cracky, G9330 100 ohm in thermal chamber (18-40c very slow ramp) over 2002-6, PT1000 Honeywell HEL-705 in G9330-10KZ over 2002-4 and finally Fluke 5450A resistance calibrator 10Kohm output over Advantest.

It drifted +8ppm already over last 12 hours. 

[Pipelie]

uploaded the A03 firmware.

[z01z]

uploaded the A03 firmware.

Thanks. Is the smaller flash still the same (A02)?

It looks like one byte was changed (apart from stepping revision numbers).
It also looks like that there's no checksum.

[Mickle T.]

It looks like one byte was changed (apart from stepping revision numbers).

This byte is a part of 32-bit floating point number, used in calibration math procedures.