EEVblog #731 - Keithley DMM7510 7.5 Digit Multimeter Teardown

[rf-design]

I would guess that the analogue supply is using a sine wave drive, probably at a fair amplitude and able to give a low distortion output. That will give low noise ( aside from the fundamental) and will provide enough power using the simple 2 turn primary and probably a resonant secondary with some fast switches as synchronous rectifiers on the secondary side. 1MHz as a drive looks about right for the size, and you will get a fair power as well.

I think the 21V rails are used for the relay drive, as they probably want to reduce the current in the coils, and a 24V coil will operate reliably with 21V, and still switch fast enough. There must be some switchers on the supply board though, as getting 5V off 21V with a linear regulator will involve some cooking. The converters are probably on the other side of the board, shielded by the transformer shield, and using those large collection of Nichicon and polymer caps as final filtering.

The isolation is mainly done by using coaxial on the primary. I will explain that:

You could imagine that a transformer have parasitic caps between primary and secondary. If there is no shielding the parasitic caps are distributed along the primary and secondary wire and couple depending on the details of t he mechanical winding. You can simplify if you assign an equivalent network for the parasitic caps with two coupling caps. If the transform is symetric both caps are equal.
If the transformer is driven by an alternate signal the rectifiying diodes connect alternating the to either one of the parastic caps. If the parasitic caps are equal and the primary driving voltage is also symmetric relative to the primary ground or case earth than the charge balance trough one rectifying period is zero. If either the driving or the parasitic cap is imbalanced you get a effective AC noise current source between the grounds. One method is to shield the driving because a symmetric drive or a balanced coupling is more difficult.

[rf-design]

Never thought I would as about Nyquist sampling on a DMM. With a sample rates as high as 1 MSps has any one thought about doing some testing see the frequency performance of the DC measurement. Haha never thought I would as that about a DMM. 18 bits is almost pushing the limits of 32 bit floating point math for a FFT.

1 second FFT with 1MSps give 20 additional bits in the mantissa which is more than 32Bit IEEE Floating allows. But 64Bit IEEE Floating could handle that. But that I expect that also with 32Bit the numerial errors are much less than the noise from the hardware. For less numerical errors you could also switch to DFT instead of FFT which less sensitive to rounding errors.

[rollatorwieltje]

That 3904 could be used as a temperature sensor. Came across a setup like that when looking at the LTC2990 datasheet, an i2c voltage/current/temperature ic.

[teddy529]

Really great tear down! Thank you for that.

Nice to see some German made WIMA capacitors on the board as well.
Now it would be nice to have a side by side comparison between this one and the Keysight 34470A 7 1/2 digit meter.

If someone is going to do it, that'd be my best birthday gift. Not just tear down, but real applications.

[Dr. Frank]

No way, guys!

Even though Wayne C. Goeke, designer of the HP3458As ADC, is the person to join Keithley, the 7510s ADC by no means comes near the 3458A!
Lay their specs side by side, and you will promptly see a big quality difference, yet.

Its linearity is mediocre 1..2ppm, again, like most all other 6 1/2 ... 7 1/2 DMMs, and still no other DMM has got this unique , true AUTOCAL function...
Obviously, the true value of this function (i.e., a complete cal lab built inside) is still not clear to everyone...

Frank

Do you know what is the specific feature which take the linearity from 2ppm to 0.2ppm in the ADC?

Due to the description in hpj 4/89, the linearity is due to complete compensation of charge injection by the FET switches, which steer the ramping currents.
It's a mix of hardware, i.e. these specially designed multi current stages with adapted FET sizes, plus a special software state machine steered integration pattern.
The 1281/8508 DMMs have also a very linear AD, 0.1ppm, but I can't tell, what's unique about these.

Maybe patents still guard these designs.

Multislope III, IV, and also the 7510 AD look very simple, a few switches, 2 ..3 OpAmps, and that's it..
You can get high resolution out of if it, but w/o further effort as above, linearity stays mediocre

Frank

PS: The 3458A has 0.02ppm typ. linearity, not 0.2ppm!!

[Dr. Frank]

That 3904 could be used as a temperature sensor. Came across a setup like that when looking at the LTC2990 datasheet, an i2c voltage/current/temperature ic.

I didn't pay special attention to these circuit elements in the video, but I recognized also this bunch of discrete transistors, coming in pairs, at it seemed to me.

Reversed bipolar transistor diodes (BC diode, afair) may well serve as ultra low leakage protection diodes for overvoltage..

There's a very uncommon diagram in AoE, about transistor characteristics, showing this low voltage range, with a comment fitting that idea..
I think, the 2N3904 served especially well for that, i.e. leakage in the fA range. (*)
Frank

AoE, p. 294-295, ch. 5.2.2, about protection of OpAmp input stages.
Figure 5.2 shows 10fA forward current at 1mV for the 2N3904 B-C diode, but 10mV for the B-E diode.

reverse currents are even smaller..

great.. no need to use expensive low leakage didoes.

[Marco]

Maybe patents still guard these designs.

Not unless they submarined the hell out of the patents.

I wonder if it wouldn't be smarter of instead of switching in the voltage-resistor combos with JFETs to switch current into the integrator. If you first create the current you can switch in voltages with MOSFETs with negligible on state resistance, no need to be in a hurry. Then you can use a single JFET, with diodes in front of it to dump the current into ground when it's off, to switch the current into the integrator (or maybe that's patented, I haven't looked).

[mikeselectricstuff]

I just dropped by this thread to ask what that guy is.  Never seen a SSOP-80 before.  They are calling it "MilliPaQ".  Good name.  I want one on my project just because it looks cool.   

I'm almost tempted to order one just to x-ray it to see what the leadframe looks like...

[Mickle T.]

I wonder if it wouldn't be smarter of instead of switching in the voltage-resistor combos with JFETs to switch current into the integrator.

Keithley 2002, Prema 6048, Advantest 6581 and many other DMMs do the same thing.

[Marco]

I wonder if it wouldn't be smarter of instead of switching in the voltage-resistor combos with JFETs to switch current into the integrator.

Keithley 2002, Prema 6048, Advantest 6581 and many other DMMs do the same thing.

That diagram still has a high speed switch (ie. non negligible resistance) for switching in the input voltage/resistor. They also have multiple high speed switches for the current sources, all build with different active components.

What I meant is that you could switch voltage/resistor combos into a single virtual ground for a single opamp current source, with slow negligible resistance switches (ie. <100 mOhm MOSFETs). Once the current stabilizes you put it into the integrator at the start of a clock cycle with a single high speed switch, when you want to change the current you take the current off at the end of a clock cycle and rinse and repeat.

[splin]

Dave mention it's 18-bit ADC, but I remember Keithley marketing materials says 32-bit? Can find source yet, but somehow that stick into my mind.

Custom 32bit Keithley special for slow measurements, 18bit AD7982 for faster readings.
I actually spy two AD7982's on the board.

Those 18 bit 1MSPS AD7982s aren't cheap devices at approx $23 each @ 1K. Anyone got the faintest idea why they would have 2 of them?

I can't think of any good reason - they *could* use them to provide full rate 1MSPS digitizing with alternating autozero and gain calibration to avoid drift during long sampling runs but it seems very unlikely. In any case they could have used a single 2MSPS 18 bit AD7641 at $29. Also the manuals/specs don't suggest it does this. 

[eV1Te]

Those 18 bit 1MSPS AD7982s aren't cheap devices at approx $23 each @ 1K. Anyone got the faintest idea why they would have 2 of them?

I can't think of any good reason - they *could* use them to provide full rate 1MSPS digitizing with alternating autozero and gain calibration to avoid drift during long sampling runs but it seems very unlikely. In any case they could have used a single 2MSPS 18 bit AD7641 at $29. Also the manuals/specs don't suggest it does this.

Maybe it uses one for the sense-inputs and one for the standard inputs (e.g. 4-wire resistance)

[Marco]

I can't think of any good reason - they *could* use them to provide full rate 1MSPS digitizing with alternating autozero and gain calibration to avoid drift during long sampling runs but it seems very unlikely. In any case they could have used a single 2MSPS 18 bit AD7641 at $29. Also the manuals/specs don't suggest it does this.

Multislope converters want to know where approximately the slope is, to avoid saturation during run up and keep run down times small ... so they could be using one of them for that.

[Mickle T.]

Multislope converters want to know where approximately the slope is, to avoid saturation during run up and keep run down times small ... so they could be using one of them for that.

All of the eight high-end ADCs in the twelve 8.5-digits DMMs don't want to know where approximately the slope is 

[Marco]

I'm pretty sure the 3458 doesn't actually go to 120 kV ... they at the very least have a limit comparator, which is an ADC.

[Mickle T.]

3458A have a zero-cross comparator only :-(

[Marco]

You're right, they probably just assume a worst case input voltage and swing it through zero more often with more switching than otherwise necessary (and overpower the input voltage on the first slope to make sure it's of known polarity). The lower end HP/Agilent multimeters with schematics available often have a signal running off to an IC with an input called FLASH, which is kinda suggestive. Keithley has a second comparator in 2001 (thus almost certainly in the 2002 as well).

[tszaboo]

Dave mention it's 18-bit ADC, but I remember Keithley marketing materials says 32-bit? Can find source yet, but somehow that stick into my mind.

Custom 32bit Keithley special for slow measurements, 18bit AD7982 for faster readings.
I actually spy two AD7982's on the board.

Those 18 bit 1MSPS AD7982s aren't cheap devices at approx $23 each @ 1K. Anyone got the faintest idea why they would have 2 of them?

I can't think of any good reason - they *could* use them to provide full rate 1MSPS digitizing with alternating autozero and gain calibration to avoid drift during long sampling runs but it seems very unlikely. In any case they could have used a single 2MSPS 18 bit AD7641 at $29. Also the manuals/specs don't suggest it does this.

Probably to have overlapped measurement/calibration cycles. Driving high speed accurate ADCs is a real task, and you cannot switch every other measurement. Settling time of 0.5uS 5V with 18 bits, and to have anti-aliasing filtering... well that is a bunch of high speed amplifiers (fairly precision) after each other.

[splin]

Those 18 bit 1MSPS AD7982s aren't cheap devices at approx $23 each @ 1K. Anyone got the faintest idea why they would have 2 of them?

I can't think of any good reason - they *could* use them to provide full rate 1MSPS digitizing with alternating autozero and gain calibration to avoid drift during long sampling runs but it seems very unlikely. In any case they could have used a single 2MSPS 18 bit AD7641 at $29. Also the manuals/specs don't suggest it does this.

Probably to have overlapped measurement/calibration cycles.

Yes, but as I said, "it seems very unlikely". Having looked at the reference manual I found the statement:

"Digitize function do not support autorange, autozero, or auto delay."

I very much doubt that it will  keep correcting the gain calibration if it doesn't autozero. In any case the AD7982 has a typical gain error temperature drift of only 1uV/K or .2ppm (and a zero drift of .5uV/K). The front end attenuators and amplifiers and filters will add some drift but probably not much. Also the Digitize Voltage accuracy 90-day specs are pretty undemanding at only 110 (measurment) + 75ppm (range) with a (best) temperature coefficient outside of +/- 5C of 10 + 20ppm.

So the question remains as to why they included the second AD7982 given that it probably costs as much, if not more than, all the components in the main multi-slope converter?

Driving high speed accurate ADCs is a real task, and you cannot switch every other measurement. Settling time of 0.5uS 5V with 18 bits, and to have anti-aliasing filtering... well that is a bunch of high speed amplifiers (fairly precision) after each other.

True, but I don't think that's the issue. And the digitize function is not exactly high precision - that 110ppm + 75ppm  (range) accuracy is barely better than 12 bits! I guess that Keithley didn't see accuracy as being particularly important for this feature or, perhaps more likely, the specs are very conservative.

I guess we need someone to reverse engineer the signal paths to the two AD7982s. Or a Keithley insider!

[PolarisAT]

Hi everyone,

first of all I really like Dave's show and the teardowns! Keep up the good work Dave!

But in this teardown I just don't get it how this special transformer/powersupply with the coax cabel works. Can someone please explain me what Dave is talking about and probably make a small drawing for me? Or even better, could Dave make a separate video on this topic? That would be great!

[plesa]

Interesting comparisons Keysight 34470A with Keithley DMM7510 http://imgur.com/gallery/Wbx7q

[SharpEars]

Interesting comparisons Keysight 34470A with Keithley DMM7510 http://imgur.com/gallery/Wbx7q

I wish there were more comparisons of the DMM7510 vs 8 1/2 digit meters for noise levels. I think it would handily beat many of them also!

[Mickle T.]

I think many of the old good nanovoltmeters beats the 7510
EM N31, Tinsley 6045 have a ~100 picovolt p-p noise. How about the 7510?

[splin]

About that coax transformer - for those who haven't already come across this. Some while ago in the "T.C. measurements on precision resistors" thread, Dr Frank mentioned a CERN paper by John Pickering et al: 'Design and Evaluation of a 10-mA DC Current Reference Standard'.

http://cds.cern.ch/record/643294/files/cer-002399331.pdf

Having recently read it took a while before the penny dropped as to where I had come across a transformer just like the one in the paper.

VI. DESIGN FOR ISOLATION

The  back-to-back  calibration  configuration is shown in Fig. 3 and inspection shows that if the power supplies generate common mode current, this will flow through the sense resistor. It, therefore, constitutes an error if it is dc or a disturbance if it is AC. A new dc–dc converter using a patented double-screened transformer construction was chosen, together with a slew-limited low-noise switching controller. This ensures that there will be no difference between use with external power supplied and use with the external supply completely disconnected when under battery power. The double-screened transformer uses two toroidal cores, independently wound, one each for primary and for secondary. Each is fully screened within injection moulded conductive plastic shrouds and coupled with an external “shorted turn” winding around both cores. The resulting coupling capacitance from primary winding to secondary screen is less than 0.1 pF and in use 10 nA of pk-pk AC coupling current is achieved compared with at least 1000 times more in most commercial units.

So that's where Keithley got the design from, and helpfully for them, out of patent by now.

[zlymex]

......
The white (ceramic ?) collar you see around the LTFLU, seems to be the exterior heater, it's named 'H2', obviously.
Its temperature might be controlled by the 2nd OpAmp , maybe.

The plastic or metallic shield around the whole assembly, also the one on the bottom side, also serve as a thermal isolation, so that the whole interior may heat up .. there's a LM35D temperature sensor also, maybe to measure if the whole assembly has been stabilized.

On the bottom side, I assume that there are 4 stable, (leaded?) resistors, which are always needed for the topology of this Reference Amplifier, aka LTFLU / SZA263 / T.I. device from all Fluke calibrators and voltage standards.

Sorry to bring this up, but this is the first time for me seeing this thread and I'm trying to clarify might-be misunderstanding regarding the reference heater.

My guess is, two N95 transistors(FCX495) are the heater drives and also function as heating elements, their emitters are connected to heating resistor R250, R499, R500 and R252 thru thick traces.
LM35D would be the temperature sensor of the oven.
Also, this part of the board/oven is a 7V reference only, so there is no precision resistor pairs to step up to 10V.