[SOLVED] High Impedance, Robust, Analog input stage for 20bit A/D

[Jester]

I'm looking for an Op amp scaling circuit that would convert a +/- 35Vdc signal to the input requirements of a Maxim 11208 20bit A/D (0-3V). The input impedance should be 10M or greater. Accuracy goal is +/-1mV or better

The MAX11208 will be powered from 3.3Vdc

Can someone suggest an appropriate OP Amp and circuit to scale this signal.

https://datasheets.maximintegrated.com/en/ds/MAX11208.pdf

I'm not married to this particular A/D, if there is a better choice available. This part was chosen for the following reasons:

• Sigma-Delta will simplify software filtering requirements
• 2 wire interface that is easier to optically isolate than I2C
• Low cost

One downside is the need for an external reference, that will add a bit to the cost.

 

[FvM]

I believe you know that +/- 35V is no voltage range for a standard OP.

So you have essentially these options:
- use an expensive high voltage OP
- design a discrete buffer amplifier
- use a passive voltage divider, similar to an oscilloscope input

 

[KlausST]

Hi,

I wonder why you want to use a 20 bit ADC, but your goal is to achieve 1mV accuracy.
A 16 bit ADC is sufficient, and usually cheaper.

You should say what the purpose of the circuit is...what is the input? Should it be a handheld device with just two cables?
Or is it installed inside a device an has fixed connections to something.

Also you should give an input frequency range...

Ask yourself how much money you want to spend, or what resolution, accuracy or precision you really need.

Some calculations:
Let's say you use an 10M input resistor fed to an inverting Opamp with 470k feedback, plus Vref/2 offset curcuit. All single ended.
10M and 1mV....if you really need that DC accuracy, then select an Opamp with less than 35uV input offset voltage and less than 100pA input bias current.
And all resistors and the Refrence should be tolerated to 0.003%.

In my eyes this often makes no sense. The cost for these devices is too high.

I'd go for low drift opamps and low drift reference and resistors. The expected overall drift should not exceed your max tolerable errors.
All the remaining gain and offset errors can easily cancelled with software.

Decide what accuracy you really need, and decide what accuracy is "nice to have".

Klaus

 

[Jester]

I believe you know that +/- 35V is no voltage range for a standard OP.

So you have essentially these options:
- use an expensive high voltage OP
- design a discrete buffer amplifier
- use a passive voltage divider, similar to an oscilloscope input

Would this circuit work?


The op amp is pretty inexpensive $1.44
https://datasheets.maximintegrated.com/en/ds/MAX9636-MAX9638.pdf

Can you tell me more about the passive oscilloscope input circuit?

- - - Updated - - -

Hi,

I wonder why you want to use a 20 bit ADC, but your goal is to achieve 1mV accuracy.
A 16 bit ADC is sufficient, and usually cheaper.

You should say what the purpose of the circuit is...what is the input? Should it be a handheld device with just two cables?
Or is it installed inside a device an has fixed connections to something.

Also you should give an input frequency range...

Ask yourself how much money you want to spend, or what resolution, accuracy or precision you really need.

Some calculations:
Let's say you use an 10M input resistor fed to an inverting Opamp with 470k feedback, plus Vref/2 offset curcuit. All single ended.
10M and 1mV....if you really need that DC accuracy, then select an Opamp with less than 35uV input offset voltage and less than 100pA input bias current.
And all resistors and the Refrence should be tolerated to 0.003%.

In my eyes this often makes no sense. The cost for these devices is too high.

I'd go for low drift opamps and low drift reference and resistors. The expected overall drift should not exceed your max tolerable errors.
All the remaining gain and offset errors can easily cancelled with software.

Decide what accuracy you really need, and decide what accuracy is "nice to have".

Klaus

This A/D is available in a 16 bit version, 2^16/70 ≈ 1mV, the 20bit version is about $1 more and I can live with the extra $1

This would be a fixed system with 2 input leads for measurement (think limited range DVM) , however the data does not simply go to a display it's sent to a uController that is also measuring other parameters. This input will be optically isolated from the uC.

The input signal will be DC, so the frequency response can be quite limited. The polarity of the input signal is unknown hence bi-polar input capability. The nominal range is ±30Vdc, however the circuit needs withstand normal ESD and should not fail if inadvertently connected to say 100V.

I really need 1mV accuracy, calibration will be done in software, and temperature range is limited to 15-30 °C of actual use. The uC measures temperature so it can compensate for resistor drift as long as its predictable.

Is there a reason I can't use a less accurate reference, say 0.1% and calibrate out the error?

 

[FvM]

Would this circuit work?

Basically yes, but you need a much better OP for 1 mV input referred accuracy.

Voltage divider ratio is about 20:1, so something like 50 µV would be appropriate. Static offsets can be removed by adjustment, but at least the offset drift should be lower.

 

[KlausST]

Hi,

Yes, the circuit is o.k.

Instead of one resistor to 1.5V, i'd go for two identical 470k, one to Gnd, the other to VREF of the ADC.
AND use two identical paralleled 470k as feedback. In total you need 4 x 470k resistors.
With the four identical resistors you decrease thermal drifts drastically.

Klaus

 

[Jester]

Basically yes, but you need a much better OP for 1 mV input referred accuracy.

Voltage divider ratio is about 20:1, so something like 50 µV would be appropriate. Static offsets can be removed by adjustment, but at least the offset drift should be lower.

How would this circuit behave with common mode voltage outside the power supply range?

Perhaps this would be a more suitable opamp (OPA376) https://www.ti.com/lit/ds/symlink/opa2376.pdf

- - - Updated - - -

Hi,

Yes, the circuit is o.k.

Instead of one resistor to 1.5V, i'd go for two identical 470k, one to Gnd, the other to VREF of the ADC.
AND use two identical paralleled 470k as feedback. In total you need 4 x 470k resistors.
With the four identical resistors you decrease thermal drifts drastically.

Klaus

That's quite helpful, I can use a 4x array

 

[FvM]

How would this circuit behave with common mode voltage outside the power supply range?

Obviously, the OP common mode voltage range must be satisfied. The positive input terminal is restricted to about +/- 30V. If you want increased common mode range, you need higher OP supply voltages or a larger input attenuation ratio.

Perhaps this would be a more suitable opamp (OPA376)

Yes, something like this.

Regarding 1.5 V bias voltage, it's appropriate for MAX11208, because Ain- would be connected to the same bias voltage. Respectively, there's no problem of a certain drift of the 1.5V source.

 

[KlausST]

Hi,

That's quite helpful, I can use a 4x array

Arrays often are no precision resistors.
But you still need "low drift", otherwise you will see gain errors. (They change their value referenced to the 5M resistors).

Klaus

 

[Jester]

Hi,

Arrays often are no precision resistors.
But you still need "low drift", otherwise you will see gain errors. (They change their value referenced to the 5M resistors).

Klaus

For the input resistors: Vishay has inexpensive 50ppm resistors and the really cheap arrays are 100ppm. Considering that the temperature range for this device is limited to 15-30°C and the uC will measure temperature and compensate for resistor drift within this range, do you think I can get away with these resistors (software is free when the code is complete)?

Caddock offers some really low drift resistors, but they are really expensive.

Thanks,
J

 

[d123]

Hi,

With my limited experience, and no empirical proof to back up remark, as I guess I miss a lot of the finer details of circuit operation: Vishay 50 and 75ppm resistors seem to work well in circuits. But that's through hole wirewound. I'm trying TE metal film 100ppm, as they are good price for 1/4 and 0.6 Watt.

OPA376: I'm bad for simulators (more time looking for missing connections that look wired!!!! grrrrrr!!!!! than checking simulations), so probably irrelevant as will have made mistakes, but comparing OPA376 and INA826 a few weeks ago, the 376 had more linear gain, the 826 drifted off the line at lower and higher ends of input range, may have been how I was trying to use them, so just ignore me.

 

[bnevins]

Try a chopper opamp setup as an inverting amplifier. The input currents are very small, so you can use large resistors.

 

[schmitt trigger]

Measuring +/-35V with a resolution of +/-1mV slightly exceeds the capability of a 16 bit AD. But it doesn't mean that you require a 20 bit AD, you may want to use a 18 bit AD, and with the money you save spend it on a good opamp.
For microvolt offset, nothing beats a chopper stabilized opamp. However the last time I looked -and it was a while back- there were no 3V chopper stabilized opamps with RR outputs.

Having said that, the major semiconductor companies are continuously coming with new, improved products, featuring performance specs that we could only dream about a few years ago.

Lastly, and this goes without saying.......a good board layout is as important as a premium device with high performance AD circuits. Depending on the complexity of the ancillary circuits, you may require a 4 layer board.

 

[KlausST]

Hi,

AD8551 or AD8571 are chopper with 3V RR.

I´m curious where one needs +-35V and an accuracy of 1mV. (for sure there will be applications)
I´ve seen such specifications, but in the end the very most of them are just "nice to have".

Even if one needs it. How can you ensure the calibration (linearity over voltage, time and temperature).
With a temperature compensation you need to calculate/compensate both offset and gain. For linearity compensation you additionally need a lookup table.

Klaus

 

[Jester]

Hi,

AD8551 or AD8571 are chopper with 3V RR.

I´m curious where one needs +-35V and an accuracy of 1mV. (for sure there will be applications)
I´ve seen such specifications, but in the end the very most of them are just "nice to have".

Even if one needs it. How can you ensure the calibration (linearity over voltage, time and temperature).
With a temperature compensation you need to calculate/compensate both offset and gain. For linearity compensation you additionally need a lookup table.

Klaus

Regarding the accuracy, I agree however this is the specification I'm working to, who knows perhaps they will verify my work with a Fluke87;-)

I found an old board I did with an AD628 and similar circuit, I just tacked on the 4M7's to the input and its surprisingly stable as is with 1%, 100ppm resistors. Sitting on my bench connected to a 0.003% standard, drift over a few hours after the initial 10 minute warm up is 0.0004%

 

[KlausST]

HI,

Fluke87V voltage measurement accuracy: ±(0,05 %+1)
--> even if it had an +/-35V range this means +/- 17.5mV.. you need a reference measurement tool 20 times better than FLUKE87.

drift over a few hours after the initial 10 minute warm up is 0.0004%

0.0004% from what?

If i calculate right, then 100ppm is 0.01% this i per klevin or per degree celsius.

15..30°C (don´t you expect any internal heating, or just somebody holds it in his hands, or the sun shines through a window...?)
means 15 x 0.01% = 0.15%
multiplied with the range of +/- 35V this gives: +/-52.5mV. Far away from your desired 1mV.


Klaus

 

[Jester]

±

HI,

Fluke87V voltage measurement accuracy: ±(0,05 %+1)
--> even if it had an +/-35V range this means +/- 17.5mV.. you need a reference measurement tool 20 times better than FLUKE87.

0.0004% from what?

If i calculate right, then 100ppm is 0.01% this i per klevin or per degree celsius.

15..30°C (don´t you expect any internal heating, or just somebody holds it in his hands, or the sun shines through a window...?)
means 15 x 0.01% = 0.15%
multiplied with the range of +/- 35V this gives: +/-52.5mV. Far away from your desired 1mV.


Klaus

Klaus, you missed my wink ;-) beside Fluke 87.

Meaning to say whoever wrote the specification might just use an instrument that is not capable of testing to the specification.

Regarding 0.0004%, after a few minutes of stabilizing after power-up the AD628 circuit (with 100ppm resistors) only drifted 10uV over a few hour period from 2.54611 to 2.54612Vdc. The source I tested with is pretty stable, and the temperature here must be also be fairly constant.

Sorry I did not clearly state the specification.

Input Range: ±35Vdc
Resolution: 0-9.9999 (100uV), 10.000-35.000V (1mV)
Accuracy: 1mV/V over temperature range


So <35mV error @35V would be acceptable, I hope to do much better by measuring the temperature in the vicinity of the measurement circuit an compensate.

 

[KlausST]

Hi,

only drifted 10uV over a few hour period from 2.54611 to 2.54612Vdc.

this is in the range of thermocouple effects. Hard to avoid.

So <35mV error @35V would be acceptable,

This accuracy should be possible with reasonable effort.
And a resolution of about 1 mV is possible too.

I´ve got the wink with the Fluke. But the Fluke is not a bad handheld one. I think 90% out there would be glad to have a Fluke87 (and know how to handle it).

Klaus