Simple DC AFE for ADC chips with unipolar diff inputs

[iMo]

I've been searching for a simple DC AFE for a basic "Voltmeter", based on ADS12xx or LTC2xxx chips. Having some ADC chips in my junkbox I want to start to play with them as fast as possible, where the target is "6 digits voltmeter only" at this stage. The approach is KISS.

Thus the basic spec for the AFE:

1. input voltage +/- 20V max
2. input impedance >>100Meg
3. DC measurements only
4. basic input protection
5. AFE not "metrology grade" one, but it should not introduce too much noise and drift
6. supporting up to 6.5 digits with appropriate ADCs (ie. see TiN's ADS1263 review and test)
7. AFE shall feed the ADC chips with differential inputs, but unipolar voltage range (0-5V)

The basic idea is to have a chopper opamp, HV one for example.
The AFE and ADC will get floating power supplies, and the ADC's INP_N input is the Voltmeter's "ground terminal".

See below the basic concept.
When switching FB to A or B you can get x1 or x10 amplification.
When switching C to A or B you can get x1 or x0.1 attenuation.

Open issues: how to protect the input of the ADC's INP_P and INP_N.
Any hints welcomed.

PS: let us not discuss the suitability of various monolithic SD/SAR 24bits++ ADCs for "DMM" purposes here, plz.

[jaromir]

Using LTC2057 to generate half of Vref seems to be overkill here. You can use something like MCP6001 powered by power supply of ADC.
JFET seems to be common choice for overvoltage protection, for its low leakage. You may need to manually select individual transistors, though.
Also, be careful with C4 and C5 selection.

[MasterT]

Emitter's junctions are "no go" anywhere near signal path.
It's not clear, why AFE is necessary, seems ADC has internal PGA with 1 G impedance and 40 M with  bypass. Why not use a simple resistive divider to extend input range?

[iMo]

@jaromir:
TI's appnote sboa058 says Siliconix's 2n4117 claims 60fA.
Tried with MCP6021 (6001 does not simulate here).
Simulation shows a little bit worse result (1.5uV off with +/-20V input).
@MasterT: it depends on ADC chip used, the goal is >100Meg (with careful build).

[MasterT]

I had negative experience using AZ OPA MCP6V28 for buffering reference voltage for 16-bits DAC. The problem was with  a bias current, I mistakenly didn't follow microchip recommendation to  keep input impedance seen by OPA equal:

Make the resistances seen by the inputs small and
equal. This minimizes the output offset caused by the
input bias currents.

  p.24

Looking at LT2057 ds, I see that LT recommends just exactly opposite:

Injection currents from the two inputs are of equal magni-
tude but opposite direction. Therefore, input bias current
effects due to injection currents will not be canceled by
placing matched impedances at both inputs.

  p.19

Doesn't matter if it's in the same direction or not, but I'd not use AZ in any high input (100 M) impedance application. On the same page 19:

When these small
current pulses, typically about 0.7nA RMS , interact with
source impedances or gain setting resistors, the resulting
voltage spikes are amplified by the closed loop gain.

V = R x I = 100 M x 0.7nA = 70 mV. 
It makes AZ even worse than LM358 with 10 mV offset, and LM358 is preferable since offset is pure DC no HF glitches, and easily subtracted by uCPU at start-up re-calibration subroutine.

[Kleinstein]

For the input protection I would consider bootstrapping the diodes. With the input amplifier as a buffer only this should be simple. It does not need that much secondary protection (directly before the TC2057) than - so some 1-5 K should be sufficient. The capacitors at the input also look like way to large - more like 100 pF-1nF would be a reasonable range.

The LTC2057 can have quite some input bias current. So it might need section or bias compensation to get really low values.
The bias cold also be a problem if just a resistor (e.g. 100 K) is used for protection. 100 pA at 100 K gives an offset of some 10 µV.

What ADC do you have in mind with differential inputs, but only unipolar range ?
This is an unusual combination. The more typical one is differential inputs and a bipolar range (e.g. LTC2440, LTC2410), or ground based and unipolar (e.g. LTC2400).

Using the additional offset of U_ref/2 adds extra errors, that could be avoided with an ADC with differential bipolar input instead.

[iMo]

@MasterT: 2057 chopper freq leakage - a good point. I saw a discussion here telling how the 2057' uA chopping currents hammer with 100kHz into the LTZ1000's crystal lattice.
There are other HV opamps, ie OPA445, with 10pA inp current, but rather large offset tempco (10uV/degC). The offset itself is not a big issue, imho, but its tempco is.

@Kleinstein:

What ADC do you have in mind with differential inputs, but only unipolar range ?

Unipolar range means the both ADC's INP_P and INP_N voltages must always be within GND..AVCC range (none of the differential inputs can go negative against GND).
I think the vast majority of monolithic ADCs (TI, ADI, LT) are of that nature.

Using the additional offset of U_ref/2 adds extra errors, that could be avoided with an ADC with differential bipolar input instead.

Yes, we do use ADC with differential inputs, therefore the additional Vref/2 offset errors cancel.
We do need that Vref/2 offset as the most ADCs do _not_ have true bipolar inputs.

PS: the second opamp creates Vref/2 - that is the virtual ground for the AFE. That virtual ground is the GND of the first opapmp. Therefore the INP_N of the ADC is kept at Vref/2, and the ADC's INP_P goes from 0..Vref for -20/20V at Vinput (in my schematics the Vref is 5V).

[iMo]

Added 2 relays. Still the INP_P needs to be protected.
Also - the ADC's diff inputs shall see small impedance, that is not the case with INP_P now..

[MasterT]

I see that 32-bits ADC doesn't require decade range switches, except one to divide 20V down to 2V.
But what about initial accuracy /tempco of those two resistors, who is gonna watch and calibrate them?
And here comes necessity to close calibration loop at the inputs, setting couples DPST to break external connection and apply GND & Ref directly to OPA. And this removes demand for AZ kind of OPA, any OPA will works. uCPU could do calibration FS and zero as often as 1 Hz. Luckily temperature is not varying too fast.
 

[iMo]

Thinking to use LT5400-8 (9k/1k or 18k/2k) or LT5400-3 (100k/10k) for the divider.
I do not know whether 20V/10k (/20k) is not too many milliwatts for the -8 chip.
Keen on avoiding clicking with relays at 1Hz, though
And yes, targeting ADS1263 and LT2500-32..

[iMo]

With ADA4522 chopper (HV type with +/-27V Vcc).
ADA seems to be a more advanced arch than the LT2057.
Simulation of a temperature sweep with worst case (?) resistor's tempcos.
The LT5400-8 tc divider mismatch is 0.5ppm/C.

[David Hess]

If you bootstrap the operational amplifier's power supplies to its output, then a low voltage part can be used as a buffer over a +/-20 volt or higher input range.  In addition, this removes error from limited common mode rejection which will otherwise be significant.

There is nothing wrong with using junction transistors for low current input clamps other than having to quality or test them however a relatively inexpensive 2N4117/8/9 JFET is guaranteed to have an input bias current of 10 picoamps or less at 20 volts.  A 2N4117A/8A/9A will be 1 picoamp or less.

[iMo]

This is with bootstrapping the OPAMP.
As an example below the low-voltage ADA4528 (5.5V Vcc max, set to 5.1V in the simulation) in use.
The ratio of the base resistors gives the voltage spread at the opamp's Vcc.
Also the resistor's values have to be such the transistors feed enough current into the opamp.

PS: this wiring is prone to oscillation with ADA4528, however. Further analysis is required.

[David Hess]

It is a good idea to minimize load changes on the output of a precision operational amplifier.  Self heating creates offset voltage changes which limit open loop gain and common mode rejection.  In this case this will also manifest as poorer linearity.

[iMo]

Do you mean to use fets in the bootstrap?
The divider there - what to use? - a capacitive divider ?

[David Hess]

Do you mean to use fets in the bootstrap?
The divider there - what to use? - a capacitive divider ?

Buffer the output of the precision operational amplifier to drive the bootstrap circuit which currently as shown has a parallel load resistance of about 10k which is not particularly low but every bit helps.

[iMo]

This is with buffering the bootstrap, and 18k/2k divider. The larger the source impedance feeding the ADC, the worse..

[David Hess]

Many implementations work in exactly that way with an operational voltage follower to drive just the bootstrap circuit.

[iMo]

And these are the simulation results with above v7.
Note:
1. source resistance of the input Voltage source is 1Meg
2. optimal resistor's tempcos with the two dividers
3. LM741 model used for the rail splitter and the bootstrap buffer.

[iMo]

v32 with low leakage input protection..