Easy DIY 5.5 Digit DVM + Volt Ref./Cal. (LTC2400+LTC6655 / SPI uC / Arduino)

[Kleinstein]

The cheap opto couplers are slow. With a careful choice if the load side one can get about a 9600 baud signal to still work, with a little luck and teaking maybe 19200.

For higher speed there are also faster optocoupler. A classic is the 6N137.

The capacirtance between the grounds can be an issue, but it should not be only one. One often has it because of common mode signals injected by the DCDC converter.

[iMo]

Soon or later you will end up with the following architecture (has been discussed here many times already):

Bootstrapped high impedance buffer->10:1_divider->ADC->MCU->optoisolated_serial

while using floating power source, like +/-20V for the buffer, and 5V for the ADC/MCU.
The serial will send voltage values, and receives some config commands only.
10:1 divider for example the LT5400 2x 9k:1k.
I made many simulations here on that AFE in past (various setups).
 

https://www.eevblog.com/forum/projects/simple-dc-afe-for-adc-chips-with-unipolar-diff-inputs/msg2120179/#msg2120179

[jorgemef]

while using floating power source, like +/-20V for the buffer, and 5V for the ADC/MCU.

I was still dreaming making it portable with some lion pack.
Going for +-20V we enter the realm of mains transformer or medical grade switching PSU, or how did you made it?

[Kleinstein]

Having a higher supply for the amplifier / buffer definitely helps. It needs a little more power, but not that much.  If battery supply is used, one can use a relatively normal switched mode up converter with some care to the filtering. The bootstrapped supply can be done with a current source and normal OP-amp, at least with a relatively low power amplifier.  E.g. I have used a MCP6V66 and TL061 for a useful range of some +-12 V. With a +20 and -15 V supply and some 500 µA of current consumption.

For the configuration one would likely not use a 1:10 divider, but less, more like a 12 V range as the target for the input. Much higher gets more tricky and power hungry for the amplifier.  Besides the amplifier also the choice of CMOS switches gets more limited beyound some +-15 to 18 V supply.

[iMo]

..The bootstrapped supply can be done with a current source and normal OP-amp, at least with a relatively low power amplifier.  E.g. I have used a MCP6V66 and TL061 for a useful range of some +-12 V. With a +20 and -15 V supply and some 500 µA of current consumption.
..

Here is Kleinstein's idea modded by iMo (his schematics found in my archive..)

PS: it is a high impedance input AFE for +/-12V range, a bootstrapped opamp, with a divider 10k/(3k||50k) and with fully differential output for a max 5V ADC (aprox 5Vpp with 12Vpp input)..

[Kleinstein]

The bootstrapped (actually more feed forward here) supply should also have a nagative side. This could be done with a divider across the zener for the feedback.
Depending on the OP-amp used one would want less voltage, more like 4-5 V as quite some of the AZ OP-amps have a limited supply range.

The circuit shown in the simulation is using a driven low side. This can complicate things when using gain for lower ranges.

[iMo]

The bootstrapped (actually more feed forward here) supply should also have a nagative side. This could be done with a divider across the zener for the feedback.
Depending on the OP-amp used one would want less voltage, more like 4-5 V as quite some of the AZ OP-amps have a limited supply range.

V2

V3 (simplified)

[iMo]

With only a 5 digit target there is no need to improve on the common mode rejection or input impedance.

I first ran across this problem with the old Siliconix 4.5 digit integrating converter chipset.  Their automatic zero cycle did not correct for common mode rejection of the integrated CMOS input buffer, which could have been worse than 72dB when 86dB was needed, and I guess nobody realized in time that this would ruin the accuracy.  They released an update which brought out the signals after the input multiplexer so that an external JFET precision operational amplifier could be used.  Precision JFET parts were better, but could barely meet the requirements back then.  They recommended the Analog Devices AD542 which was about the best available part at the time.  These days there are lots of outstanding JFET parts which can handle 6 digits, and 7 digits with grading.  The OPA145 series amazes me.

..LD120/121A/122 I would assume.. Great app notes with very details you may find in the Siliconix 1982 databook.. I got 120/121A in my collection (except that I have a working DMM based on that chipset), perhaps it would work 5.5 digits with discrete chips instead of the 122 and a tiny fpga (instead of the 121A).

[David Hess]

With only a 5 digit target there is no need to improve on the common mode rejection or input impedance.

I first ran across this problem with the old Siliconix 4.5 digit integrating converter chipset.  Their automatic zero cycle did not correct for common mode rejection of the integrated CMOS input buffer, which could have been worse than 72dB when 86dB was needed, and I guess nobody realized in time that this would ruin the accuracy.  They released an update which brought out the signals after the input multiplexer so that an external JFET precision operational amplifier could be used.  Precision JFET parts were better, but could barely meet the requirements back then.  They recommended the Analog Devices AD542 which was about the best available part at the time.  These days there are lots of outstanding JFET parts which can handle 6 digits, and 7 digits with grading.  The OPA145 series amazes me.

..LD120/121A/122 I would assume.. Great app notes with very details you may find in the Siliconix 1982 databook.. I got 120/121A in my collection (except that I have a working DMM based on that chipset), perhaps it would work 5.5 digits with discrete chips instead of the 122 and a tiny fpga (instead of the 121A).

That is the one.  If you check Google carefully enough, you will find that I have twice designed a replacement front end for these chipsets because burning out the input multiplexer is a common failure.   I designed a replacement twice because I forgot that I did it the first time.

The LD120 has the terrible integrated CMOS buffer, and the LD122 is the same chip without the integrated CMOS buffer, so that an external JFET precision operational amplifier can be used instead.  Tektronix used the AD542 for the buffer, and I have not seen any recommendation from Siliconix, but I assume there was an application note from Siliconix for the LD122 which I never found.  The datasheets for the LD120 and LD122 listed the same accuracy, which is a laugh; there is no way the LD120 could have met those specifications with an CMOS buffer uncorrected for common mode rejection.

I currently have a DM501A with the LD120 on my workbench with a bad input multiplexer, and picked up the parts to replace that part, including an OPA140 for the buffer and a DG419 for the input multiplexer.  Higher serial number DM501As used the LD122 with an AD542 which is just barely good enough.  At some point I would like to have a stack of DM501As for general use.  It might be fun to make a little board for fixing LD120/LD122 instruments, but I fear there will not be enough demand to make it worthwhile.

The run-up and run-down design should be more accurate than Intersil's dual-slope design because dielectric absorption produces proportionally less error, but Siliconix did not correct the common mode rejection of the input buffer so was worse overall.  HP published even better application notes about the run-up and run-down integrating ADCs, but for a modern design the LTC2400 or similar is better in every way.

[iMo]

You may easily download the "1982 Siliconix Analog Switch and IC Product Data Book" (24MB) where there is about 100pages dedicated to the 120/122. There are design notes for the external opamp/switch as well. They claim 1uV resolution at 20mV FS with the 122.

[Kleinstein]

The bootstrapped (actually more feed forward here) supply should also have a nagative side. This could be done with a divider across the zener for the feedback.
Depending on the OP-amp used one would want less voltage, more like 4-5 V as quite some of the AZ OP-amps have a limited supply range.

V2

V3 (simplified)

The circuit would like a capacitor in parallel to R8 to improve stability.  The ADA4528 is not a good choice for a high impedance. More reasonable OP-amp choice for the input buffer are  max4238, LTC2050, OPA333, AD8628, MCP6V76, OPA387.  For the ADC driver the ADA4523 is also overkill and not idea with 47 K resistors - more like a job for a MCP6V76. The inverter part (u3) does not even need to be an AZ type,  it only sets the common mode voltage for the differential ADC. So this could be somerhing simple like MCP6001.

....
I currently have a DM501A with the LD120 on my workbench with a bad input multiplexer, and picked up the parts to replace that part, including an OPA140 for the buffer and a DG419 for the input multiplexer.  Higher serial number DM501As used the LD122 with an AD542 which is just barely good enough.  At some point I would like to have a stack of DM501As for general use.  It might be fun to make a little board for fixing LD120/LD122 instruments, but I fear there will not be enough demand to make it worthwhile.

The run-up and run-down design should be more accurate than Intersil's dual-slope design because dielectric absorption produces proportionally less error, but Siliconix did not correct the common mode rejection of the input buffer so was worse overall.  HP published even better application notes about the run-up and run-down integrating ADCs, but for a modern design the LTC2400 or similar is better in every way.

For a DMM the LTC2400 has some limitations as it is single ended input and essentially positive voltage only. To allow positive and negative readings other more modern SD ADC chips would be a better choice. The multislope run-up of the LD120/LD122 has quite some advantages compared to the classic dual slope. The ICL710x chips just had the better front end and better auto zero implementation.

[iMo]

..
The circuit would like a capacitor in parallel to R8 to improve stability.  The ADA4528 is not a good choice for a high impedance. More reasonable OP-amp choice ..

That schematics is just an idea, simplified (you have to add the decoupling, 5V driver opamps would require handling their Vcc voltage, etc.), the opamps in the above sim are placeholders for the sim only. Would be interesting if somebody here built that AFE and characterized it with a modern 24+bit ADC..

[jorgemef]

Hi,
That AFE is for ADC with differential input, right?
If I would like to use it with ltc2400 I would need some voltage level translation, or different grounds, is this right?

[Kleinstein]

The LTC2400 is a bit odd, with only a positve sign. Most other SD ADCs come with differential inputs. It would make relatively little sense to add the level shift / differential conversion. It is just easier to use a different ADC, like LTC2410, MCP3561, ADS1219, AD7190 or a similar different type.

There is an older Hameg DMM that uses the LTC2400, but the extra polarity switching is a bit inconvenient.´

A starting point could be a configuration a bit similar to the Sigilent 3055/65.
From the pictures it looks a bit like using a buffer/ amplifier first with a circuit to generate a differential signal by driving the low side. This is a bit like the circuit IMo showed. Than come optional dividers and buffers right at the ADC inputs.

[iMo]

This is an another example of AFE (I elaborated here in past as well).
Example only, simplified, the opamps used in the sim are placeholders only..

[iMo]

For the LTC2400 you may try for example this unipolar one.
It has high input impedance (depends on the first opamp), bootstrapped, low output impedance, unipolar..
Example only.

Note: no need to have the +18V/-5V supply floating as the grounds are common. The output divider there is based on the LT5400 2x9k/1k version, you may use any such you get 0-5V at the LTC2400's input.
You may go to higher input voltages, as the input stage is bootstrapped. The zeners in the bootstrap have to be such you get required supply voltage at the opamps (like 5V max with 5V opamps). The second opamp there is a buffer deloading the first opamp. It is wired inside the loop of the first opamp, so it could be almost any opamp with some output current capability and 5V supply (in this example). The power source for the AFE has to be such the bootstrapped opamp stage "fits" inside its span considering the required input voltage range (see the bottom graph).

The overvoltage protection is bootstrapped as well under normal situation, thus there is almost none leakage current even with ordinary diodes, like 1N4148..

[David Hess]

You may easily download the "1982 Siliconix Analog Switch and IC Product Data Book" (24MB) where there is about 100pages dedicated to the 120/122. There are design notes for the external opamp/switch as well. They claim 1uV resolution at 20mV FS with the 122.

Thanks, I have been using various separate datasheets, and the 1985 databook which is missing the application notes.  I am not sure why I did not check the 1982 databook.

For a DMM the LTC2400 has some limitations as it is single ended input and essentially positive voltage only. To allow positive and negative readings other more modern SD ADC chips would be a better choice.

Hmm, I thought the LTC2400 was better than that, and now I cannot find the better LT parts on the Analog Device's selection guide.  Did AD discontinue them?

What is the current darling of high resolution and accuracy SD ADCs?

The multislope run-up of the LD120/LD122 has quite some advantages compared to the classic dual slope. The ICL710x chips just had the better front end and better auto zero implementation.

We used the ICL710x chips, and I did not find out about the Siliconix chips until after they were discontinued.  Getting 20 to 40 thousand counts out of the ICL710x chips was a real challenge because of dielectric absorption.  Texas Instruments had a flawed but competitive solution at that time also.

I think it was the old ICL710x application notes which showed an RC oscillator for timing, but we figured out that it had too much jitter causing flicker.  The Q of an LC oscillator was much better and solved this problem.

[iMo]

90% of all modern 24+bit ADCs chips are with differential inputs, but both inputs must always be within Vgnd..+Vcc. So you have to create a floating AFE such the signals are differential around a "common voltage" which has to be always positive. There are several ADC chips (afaik) which work with negative voltages as well ("true bipolar", internal charge pump, like +/-5V or +/-10V inputs), but usually not "high end precision one".

Precision 24bit SD/SARs often mentioned here - AD7177, ADS1257/63, LTC2500-32, AD4630-24, AD4030-24  (all are not true bipolar afaik).

[Kleinstein]

There are a few low INL SD ADCs, e.g. :
AD7177    1 ppm INL typ  , very low noise , but a bit expensive
AD7190    1 ppm INL typ  , used in SDM3055
AD7175    1 ppm INL typ  , used in SDM3065   up to 250 kSPS
AD7768    1.1ppm INL typ  , up to 1000 kSPS
LTC2442   1 ppm INL typ with 4 V ref.
ADS125H   2 ppm INL typ , 40 kSPS,  up to +-20 V input range  - could be a simple solution
ADS1259   0.4 ppm INL typ  14 ksps

Especially with the faster, really low noise SD ADCs the INL is not just a function of the ADC chip, but also depends on how the reference and inputs are driven. So the real life INL may end up worse or it needs quite some extra effort for those drivers. So one would still need to check if the circuit is working as linear as planed. The older LTC2400 and other more low power ones are easier to drive.

The ICL71xx sereies is really made for 3.5 to 4.5 digits and not more. The DA is limiting and also the DNL at around zero needs trim and measureing the input only for 1/4 the time is limiting.

[David Hess]

Analog Devices changed (made worse) their selection guide and it seems to keep missing parts now.  The LTC2442 was probably the one I was thinking of.

The ICL71xx sereies is really made for 3.5 to 4.5 digits and not more. The DA is limiting and also the DNL at around zero needs trim and measureing the input only for 1/4 the time is limiting.

Dielectric absorption in the integrating capacitor limits all simple dual-slope converters to about that.  I think we had to grade the capacitors individually to meet the claimed ICL71xx specifications.  The run-up and run-down converter has a big advantage here because it gets a lot more "travel" across the integrating capacitor for the same dielectric absorption.  The HP applications notes have the most detailed description of what is going on that I know of.

[iMo]

Here is a sim with the unipolar AFE and the latest AD4630-24 high end ADC.

Impedance of the 10V source is 10Meg, divider is 9k/1k, and the ADC's diff input is biased at +1V to see the difference against the output of the AFE (after the divider we should get 1V).

The AD4630 shows the output value as a "voltage", the ADC's timing comes from the ADI's example (2Msps).
Each step at the Vout (ADC data out) represents an ADC conversion, imho..
You may see the 1k divider impedance is still high, as we see couple of uV ripples there, coming from the ADC input, at least that is my current understanding..

[David Hess]

Even at zero volts, 1N4148s have a lot of leakage.  To put it another way, the change in current for a change in voltage at zero volts is still high.  I would still use low leakage diodes for D3 and D4 even though they are bootstrapped to have zero volts across them.

Maybe their leakage current is smaller than the input current noise?

Bob Pease had the following to say about it, plus the leakage increases dramatically as temperature increases.

[David Hess]

You may easily download the "1982 Siliconix Analog Switch and IC Product Data Book" (24MB) where there is about 100pages dedicated to the 120/122. There are design notes for the external opamp/switch as well. They claim 1uV resolution at 20mV FS with the 122.

Thanks, I have been using various separate datasheets, and the 1985 databook which is missing the application notes.  I am not sure why I did not check the 1982 databook.

I see from application note AN80-8 that I calculated exactly the same common mode rejection that Siliconix did - 86dB.  That is reassuring.

[iMo]

Btw you may replace the input buffer (or amplifier) in the LD120 as well.
You would need a switch (with inverter for M/Z) and an opamp.

[David Hess]

Btw you may replace the input buffer (or amplifier) in the LD120 as well.
You would need a switch (with inverter for M/Z) and an opamp.

That is what I am doing.  My DM501A has the LD120 and not the LD122, but later serial numbers have the LD122 with an air wired AD542 as shown below.  I am not sure if there is Tektronix documentation showing the changes.