Advantest R6581 8.5 digit DMM mini teardown/repair

[ramon]

Presto ! Please raise your hand whenever did you bought your R6581T *new* from authorized dealer. Anyone?

I guess the R6581T was the low cost 8.5 DMM to be use as cannon fodder for automated torture test when you don't want to risk your proven 3458A unit(s). And as eurofox said almost all second hand units have those torture test hours clearly visible on the (extremely dim) display. If some company had several 6581T units, the unit that goes out to sell is that old faulty-unreliable-broken unit. As simple as that.

If anyone bought the unit new and detected all those issues (6.5 DMM performance? WTH!), then it could have called directly the distributor or manufacturer to blame them all and ...

Life goes on ... Anyone is kind enough to summarize all previous testing and write some simple tests for dummies. Does R6581T have test points? is that documented anywhere? If someone is so kind, maybe I will be willing to help with all (our) issues. But don't know if I will be of any help as I don't have scope now.

[zhtoor]

i have asked the adcmt company for the specifications sheet for the T version, so let us keep our fingers crossed.

i agree with you, the specs may turn out to be exactly like R6581/D or maybe not, or not available at all.
i have'nt seen any calibration sticker on R6581T, has it been calibrated / tested / verified at any national lab? if so the specs should be available,
if the specs are not available at all, or are only available to adcmt as part of a bigger system, then trouble.

regards.

-zia

[eurofox]

i have asked the adcmt company for the specifications sheet for the T version, so let us keep our fingers crossed.

i agree with you, the specs may turn out to be exactly like R6581/D or maybe not, or not available at all.

regards.

-zia

Maybe you missed my last message because I edit it, I double check and when I start the instrument it indicate R6581D why?

eurofox

[quarks]

afaik R6581D is for DC only

my R6581 says R6581 and has AC function



Edit: removed AC board, now it shows R6581D as expected

[zhtoor]

i have asked the adcmt company for the specifications sheet for the T version, so let us keep our fingers crossed.

i agree with you, the specs may turn out to be exactly like R6581/D or maybe not, or not available at all.

regards.

-zia

Maybe you missed my last message because I edit it, I double check and when I start the instrument it indicate R6581D why?

eurofox

no idea, maybe the firmware used is the same and specifications are different, but it definitely looks like a dvm with a long calibration cycle.

regards.

[szszjdb]

Hi,  Mr.Kleinstein,
Thanks a lot!

Understood that the current leakaged from the control gate of the JFET switch. That must be !

I tested the ADIN with the PIN20 and PIN27(GUARD) of MUX and recorded the front of the signal period . The input voltage changed from 10v to 7v and 4v to cover the full range.  Attached FYI.

EDIT: The bias of the slow JFET is about 0.5ma tested before. And the trimmer is for the zero adjust of the slow JFET amplifier. Both seems no issues.

Best Regards,
szszjdb

[TiN]

my R6581 says R6581 and has AC function

Any chance you can dump firmware for us, please? 

[quarks]

Any chance you can dump firmware for us, please? 

no problem, I  already gave a copy to Mickle and I can send you a copy too
But afaik my firmware version is already on your Server
If not just send me a Mail.

[Kleinstein]

Thank you, szszjdb, for new curves.

They clearly show the limited slew rate.  Needing up to some 200 µs for switching and rising the voltage back looks rather slow to me. This would add a considerable delay to the measurement. So I wonder if this is normal. Besides a limitation of the amplifier, this could also be a limitation of the JFET switching: During turn on the guard driver gives a voltage that essentially follows the source voltage and thus only little higher than the gate voltage. So the current provided by the 100 K resistor to charge the gate is relatively low. It might thus take long to charge C026.

From the initial rise of the gate voltage I would estimate (C026+C_gate)*100K = 10 µs and thus around 100 pF.
So it would take some 2 V at the 100 K and thus 20 µA to get the observed 0.2 V/ µs. So it could be just a limitation due to C026.
Thus the slow switching might be just normal. Maybe too slow in some meters (e.g. due to threshold variations with the JFETs in the MUX).


I mixed up things a little in my last response. So the test with +4 V to +10 V of cause did not show the transient problem during AZ switching. This happens with a negative input. So the curves don't help much in finding the AZ problem. They just help to understand how the timing during switching works.

Leakage due to a gate biased foward during switching is at least one possibility for the extra leakage in AZ mode. However so far I can not see an indication for this.
Another possibility could be if for some reason the JFETs for DCV_in and Lo on phase overlap a little: this could happen at a very negative input voltage, as it requires the gate voltage to get very negative to turn off. Too large a value for C026 and/or R083 together with a high threshold could cause this.  A reduced value for R083 (e.g. a second 10 K in parallel) could be worth a test. The old picture NEG10_PIN17(Y)_PIN20(R)_FRONT_9.jpg might point to this direction: the connection to the input might not be totally off, when the channel Lo turns on. It's not clearly visible, but it looks like it could be close to the edge.

[Mickle T.]

Found a very strange behavior of 6581 A/D converter 
While shifting ADC zero with 898 kOhm resistor, hot connected to the V-to-I converter's opamp U200 (pin 2 - current summing point), I was able to correct INL and rollover error in 1-1.5 0.5-0.7 ppm range. The minimum error was reached at 25 uA extra current (898 kOhm, connected  to -27 V supply), which corresponds about -0.5 V ADC zero offset. Of course, with a such huge offset DMM can't start and do ACAL without error.

[eurofox]

This is the answer I got from my email request to ADC corporation ....

Thank you for your inquiry.
First of all, R6581T was made custom made model for ADVANTEST.
Therefore we can not provide R6581T to general market.
If you need R6581T and manual of it, please contact ADVANTEST sales department in Europe.
And We are provide 6581 to market. but this 6581 don't have English documents.

[Kleinstein]

The effect of added current is a little strange.  Normally just an added current to the input would correspond to an added offset and should not have a direct  effect on INL. So far my best guess for a nonlinear effect (to compensate the INL) in this case would be via a change in the -27 V supply.
Does a similar current coming from the -19 V reference (resistor around 600 K) have a similar effect ?

If the compensation is not just pure luck or tuning the resistor to get zero error at the few observed spots - it could be a change in the -27 V to cause the observed INL. Changes in the supply could be one way of getting nonlinear effect in the low ppm range - however so far I see no obvious path with poor PSRR. The small offset at the comparators inverting input (R260,R261) could be one path where the supply could have an effect, but I doubt it would be much. 

A systematic test would be measuring the effect of a change in supplies on the result and the effect of the input voltage on the supplies. By measuring in two steps one should be able to see even small contributions to INL, as usually both steps should give error attenuation. Most (ideally all) of those paths should not give much of an effect, but one could at least eliminate these possible error paths.

[szszjdb]

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

I  retested the ADIN with the PIN17, PIN20 and PIN27(GUARD) of MUX and recorded the front of the AZ period at the input of -10v , -7v and -4v  . From the photos of PIN17 vs PIN20(Photo No. 4,5,7) , it might have the change of the overlay conducted of both JFET for AZ and DCV switch.  How about remove the R083/C026 to accelerate the turn off of the DCV JFET?  Attached FYI.
The analog PCB under testing is borrowed from my friend and I can not modify any thing on it. I will try the change of R083 on my PCB later.

Would like to have your further advice!

For  the INL of ADC , there might have issue with the U214, switch of the CINT ?  How about the Ron of the switch , which might count to the DA?

Best Regards,
szszjdb

[Kleinstein]

The file * NEG10_ADIN(Y)_PIN20(R)_AZFRONT-3.jpg could show some indication for cross conduction: that initial faster rise of ADIN could be from the input connection still turning off. One might be able to see the current at the LO input (pin 10).

For a test one might be able to add a little capacitance (e.g. 20 pF range) to C26 with a probe, so without soldering. If this makes the AZ leakage worse it would give some indication.

C26 and R83 do make sense, so removing them would not be a good idea. One might however try something like a smaller value for R83 to speed up the turn off of the input. The downside could be a stronger spike at the input as the switching would be faster. It likely would not take much to prevent cross conduction so even a small change could be all it takes. Similar just a little more capacitance could cause cross conduction. So with the first board one could do a test with maybe an extra 50 pF to provoke the cross conduction. A much larger R83 would not allow the FETs to turn off at very negative input voltage.

For the ADC I don't think the R_on of U214 would be a big issue as the time constant would be rather short (e.g. 100 Ohm * 20 nF = 2 µs) compared to the time for the rundown. With U214 I would more worry about coupling capacitance that could influence the fast mode. However the fast mode is not that high resolution and INL is less of an issue here.

With a reasonable 2.nd meter one could do checks on some of the supply voltages, if they change with input voltage. The +-5 supply of the comparator could be worth checking in this respect. 

[Pipelie]

Little bird show me photo of (later?) R6581T, which have DCV 300V PK MAX Input labels.
Seems like Advantest were not able to make up mind about 6581 specs for a while. 

Video on TiN server that show that I'm not telling "fake news"

Well, I watched it, and Mr. eurofox misled us, as meter he has is 6581T , not the real 6581.  No wonder Mickle.T. did not see any difference  .

As far as I know,  the  R6581T with date code 03xx or 04xx on the ICs have DCV 300V PK MAX Input labels.

[Pipelie]

Hello guys,

I just got my R6581T last week,  and I want to know the linearity of this DMM compare to HP3458A.

Here is the result I get.

forgot to mention the PLC of 6581 I used is 100.

[Pipelie]

short-term stability verify:

source: fluke 731B, 0.1-10Hz noise is 1.4uVpp

and check the attached files for more results

[zhtoor]

hello friends,

can anyone explain the oddity in the following pictures?

-zia

[Macbeth]

hello friends,

can anyone explain the oddity in the following pictures?

-zia

You mean the custom rack mount? Perhaps they had some kind of shield to protect the inputs from draughts and/or factory workers accidentally unplugging leads and putting them back wrong? 

[szszjdb]

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

I had added a 50p COG cap on the C26, but found no obvious change with the -10v bias test at AZON mode. The time to discharge from -10v to -4v is about 1.54s for both with and without the 50p cap.  Attached the xls file, FYI.

I also tested the +5v of the U208 and it is about 4.78650v with 0.2-3mv noise on it. Is that normal?

Further  check with the new analog PCB shows that the INL is much better than mine, which has the max. around 0.47ppm in the positive range. The bias current is smaller and so the error of the totally reading to the sum of the resistor voltage become small enough to around 15uv , which is about 130uv in my old analog PCB.  It proved  that the bias of my old pcb is too large compared to the new one.
The 6581 has 2 record named with 6581-1 and 6581-2 in the picture.  Also attached FYI.

Would like to have your further advice!

Best Regards,
szszjdb

[Pipelie]

more results coming...

the linearity is not bad when using negative polarity input.

[eurofox]

more results coming...

the linearity is not bad when using negative polarity input.

Maybe a good idea is to check the voltage PPM <> when you change polarity.

The calibration sticker show last calibration in 2005 valid for 10 years 
This instrument is in 2018 not calibrated during the last 13 years   

[Kleinstein]

The negative polarity range really does not look that bad - for both the meter from Pipelie and szszjdb.
The odd thing is the kind of jump near zero and the poor INL in the positive side.  So far I don't really see a point that is changing so abrupt when going from negative to positive. There is this split in the calibrator, but there is still a difference when you look at the difference to the 3458. So it seems to be more than just the step from the calibrator.

For the INL due to R200 (e.g. self heating) there is kind of sweet spot around -2 V. So some reason to get better values for the negative side, but not that much. It would be more like a contribution symmetric around -2 V.
Otherwise I don't see much special with the range around zero: The current source at the input of the ADC is at about the middle of the range, where no abrupt change is expected. The run-up cycle feedback is about 50:50 PWM for S1024 and thus far from settling times. The amplitude for the integrator output swing would be the largest, but this would be a broad maximum around 0 and not like a step change from negative to positive.  The AZ switching would change from up to down, but it should be relatively fast in both cases.  So I doubt that there is trouble with insufficient waiting times after switching.  The rundown phase should be longer (e.g. 200 µs minimum, likely more) so that there should be sufficient time, even at high voltages where switching can get surprisingly slow.  Mickle has probably already checked the relative timing of AZ switching and rundown.


For the long time since last CAL, it depends on the applications. In some cases you may not need a calibrated meter, and with higher speed measurements the cal requirements are not that stringent as resolution limits the accuracy. The more important point would be to check those modes for something like settling time if not done properly in the self test. The odd thing is that there are usually no accuracy ratings for a 10 year cal cycle.
They seem to distinguish between an external and internal cal. So there might have been less stringent calibrations to an internal calibration (e.g. to a 10 V ref only) in between.

@ szszjdb:
The discharge curve shows really high currents at high voltage. However this is the negative sign and thus the side where INL measurement is good. So input bias does not seem to be the main source of INL error. So this might be 2 separate problems.
Depending on the BW a 3 mV of noise on the 5 V supply of U208 would be normal. The interesting point would be if this voltage for some reason changes significant with input voltage.

With the 3458 available one could do an INL test for only the ADC: use the 3458 to measure the voltage at the AD_IN test-point (TP200). The test would have to be without AZ - so it could still be a little different from the normal AZ mode results. The ADC itself uses a second internal zero mode and should thus show very little drift.

[eurofox]

This is what I get on my R6581T:

[eurofox]

For the long time since last CAL, it depends on the applications. In some cases you may not need a calibrated meter, and with higher speed measurements the cal requirements are not that stringent as resolution limits the accuracy. The more important point would be to check those modes for something like settling time if not done properly in the self test. The odd thing is that there are usually no accuracy ratings for a 10 year cal cycle.

I just wonder why you need a 8 1/2 digits meter in this case since a 5 1/2 digits is good enough if precision is not critical.
Is like buy a Ferrari to drive only in the city with maximum speed is 50KM/h 
Well at least you can use the Ferrari as a "girl catcher" but with a multimeter I don't think you can catch a girl