Suggestions needed on 24-Bit-ADC-Attenuating Differential Buffer

[Echo88]

I changed this post to state my question more precisely; maybe this will lead to more discussion:

I want to use a Differential-Input 24-Bit ADC, in this case the LTC2442 (220nVrms noise, 6.9 Samples/s, +-2.5Vin @ 5Vref), to measure a +-10Vpp bipolar input signal. The input voltage will mainly come from a Fluke 5440B-calibrator or from reference voltage designs, so no high-impedance-sources.
To not spoil the specs of the LTC2442 i want to design a fully differential input attenuator design, which has a high input impedance >= 1GR and should have the lowest INL and best stability possible. Gain/Offset-errors can be later calibrated away by a multiplexer.
After consulting my Art of Electronics book i concluded the following:
-conventional fully differential dont provide the high input impedance needed
-instrumentation amplifiers arent stable/dont have CMRR when they are attenuating rather then amplifying and they are typically single ended output

At the moment im using the PGA280, which is already good for my needs. However: the datasheet lacks the INL-specification for the used gain-modes /8 *1.375 (+-14.5Vin max @ 5Vref) and less noise would be nice.
The, now obsolete, ISL28617 also looks rather nice. Gotta study the datasheet.

I have attached the two ideas i had for such a attenuating buffer design. The first is just thought-based on LTC1043 and the second uses the following stages: buffer, attenuator, instrumentation amplifier and single ended to differential output-stage. Of course the LTC2057 and LT1167 are just exemplary and might be changed to more fitting types. Single ended to differential design: (taken from http://m.eet.com/media/1134731/15496-81904di.pdf page 5).
This link also suggests a design: https://www.edn.com/design/analog/4426097/Differential-I-O-low-power-instrumentation-amp

Maybe the suggested amplifier from page 27/28 in the LTC2442-DS is better suited for the task? I lack the necessary insight to see the pitfalls in this case.
Maybe a single-ended to differential-amplifier with external high quality resistors (metal foil/precision wirewound) yields better INL and lower noise, than the PGA280 with its internal resistors?

It would be very nice if you guys could comment/critize on the best possible design, regardless of the resulting cost if one would use vishay foil resistors.

Relevant datasheet-links:

http://www.analog.com/media/en/technical-documentation/data-sheets/2442fb.pdf
http://www.ti.com/lit/ds/symlink/pga280.pdf
http://www.analog.com/media/en/technical-documentation/data-sheets/2050fd.pdf
https://www.intersil.com/content/dam/intersil/documents/isl2/isl28617.pdf

Similar discussion: https://www.eevblog.com/forum/projects/diff-out-instrumentation-amplifier/ but the needed attenuation limits design-choices.

Thanks,

Echo

[moffy]

You might want to look at the LM4562 etc opamps which have (THD+SNR) of (0.00003%). Not sure about their offset stability but you could use a chopper amp to stabilise that.

[Echo88]

The suggested OP Amp is nice, but i rather need differential-amplifier-topology suggestions.

[branadic]

Have you considered using LTC1043 with divide by 2 in both paths?

-branadic-

[Echo88]

Thats an interesting idea! I hope to own a working LNA in a few weeks, so i can finally measure the noise of such LTC1043-based-circuits.

[Andreas]

Hello,

the problem is that the LTC1043 only can be used with max 16V (+/-8V) power supply.
For a +/-10V cirquit minimum +10/-10V supply would be necessary.
So you would have to use a different multiplexer (low leakage, low charge injection) with +/-15V supply.

with best regards

Andreas

[branadic]

Datasheet specifies a maximum supply voltage of 18V, however, not enough for +/-10V at the input as Andreas already stated.
Have you checked some meter datasheets with schematics included? Prema 5000 manuel with 1G input impedance and something like that?

-branadic-

[iMo]

+-2.5Vin @ 5Vref

Mind the Vins are from Gnd-0.3V .. Vcc+0.3V (abs max) only.

[branadic]

Using a differential ADC you won't connect them with common ground. Similar to a voltmeter you just connect the differential inputs to each other.

-branadic-

[Echo88]

Indeed, overlooked that the LTC1043 isnt capable of the +-10Vpp input. 
The LTC1043 states a unique charge cancellation:
"As shown in Figure 4, there is a predictable and repeatable charge injection cancellation when the input voltage is close to half the supply voltage of the LTC1043.
This is a unique feature of this product, containing charge-balanced switches  fabricated with a self-aligning gate CMOS process. 
Any switch of the LTC1043, when powered with symmetrical dual supplies, will sample-and-hold small signals around ground without any significant error."

If one would bootstrap the power supply of the LTC1043 to the input voltage via a very low input bias OP, it would be possible to always stay in the charge cancellation region of Figure 4 of the datasheet i think.

I gotta study the datasheets of MUXes like the MUX36D04. This device claims ~0.5pC source-drain and ~2pC drain-source charge injection-maximum over a +-10Vpp input range (Figure 12 and 13, Vin +-10V, +-15V supply). 2pC into 1µF would equate to a 2µV charge injection error. Hmm.

The Prema 5000/5001-pre-amplifier consists of a single ended ICL7650-buffer-stage/voltage-divider and Gain-/Offset/Signal-Multiplexing via "SH10", which is assume is as Sample/Hold-Stage. All other high precision DMMs that i know of use discrete JFET-Input-Stages to obtain the low noise/high input-impedance. Like described in AoE3 in the "DMM Design by the Masters"-Section.

http://www.ko4bb.com/getsimple/index.php?id=download&file=06_Misc_Test_Equipment/Prema/PREMA_5001_6001_Digital_Multimeter_w_schematic.pdf

[Kleinstein]

The LTC2442 and most SD ADCs are different from the usual multi slope ADCs in that they have a differential input and usually low offset drift and also the smaller input range. So the input circuit of the more normal DMMs does not help that much.
To get the best INL performance one needs to also keep the common mode voltage approximately (e.g. nominally, but allow for small errors) constant. So the differential input can be more of a problem than a real help.

One possible option would be to consider would be to amplify / buffer (e.g. 1 AZ OP)  the one input and drive the other input terminal to about the inverted signal. This way one would have the constant common mode voltage with very little effort and no need for well matched resistors. The higher voltage ranges (e.g. > +-2.5 V) would require a divider with 3 resistors and extra higher voltage switches before the MUX (maybe something like max388 which include over-voltage protection).  So the prime range would be +-2.5 V and a range of up to something like +-20 V could still be high input impedance, but accuracy and possibly linearity limited by the 3 resistors.