Group Actions
Engagement
  • Author Author: Jan Cumps
  • Date Created: Date Created
  • Views 1617 views
  • Likes 9 likes
  • Comments 11 comments
Related
Recommended

Measure the Analogue Front-end of Data Acquisition Board for Pi Pico

Jan Cumps
Measure the Analogue Front-end of Data Acquisition Board for Pi Pico

  designed a Data Acquisition Board ( ) for the Pico:  Data Acquisition Board for Pi Pico . In this post, I'll measure the analogue front-end, and show how it behaves within its +- 4V range.

image

I've attached probes to the main points of interest. Then applied input values between +4 V and -4 V, in steps of 0.5 V.
The data is collected in a spreadsheet to try and spot trends and stability.

Measurement points

1 5V supply power
2 3.3V supply power
3 GND supply power ground
4 AIN_1 test voltage input
5 AIN1_P opamp U1D out output
6 AIN1_N opamp U1B out virtual ground

image

Test results and calculations

  • The first 5 columns reflect the testing points of the previous chapter (without GND), and are all measured (in V).
  • diff: ainp - ainn, in essence, it is input * gain, measured
  • ain calc: diff * theoretical gain, (circuit gain, based on resistors used, is 0.38298), should ideally match ain
  • ain err: ain -  ain calc. How much does the calculated ain differ from the actual input
  • gain: what is the actual gain of U1D at each sample

image

Graphs and trends

Error

Let's first look at the error across the range:

image

This is very linear. and crosses at 0V input. That points to (discuss if you have a different opinion):

  • no offset error
  • the gain isn't 0,38298, but closer to 0.385

Here is the same graph if I set the gain for calculations to 0.385:

image

This is good news for calibration algorithms. Easy to program as a function. No lookup tables needed.

Gain

The gain is also consistent across the range. I blame the peak at 0 to calculation error factor (discuss). Everything is close to 0 at that point.
I did double-check the values though.

image

Here are all measurements:

image image

image image

image image

image image

image image

image image

image image

image image

image image

image image

And the spreadsheet:

8311.measure_analogue_frontend.ods.zip

Parents

  • Little more on accuracy: If we target 3.5 digit, or displaying from -4999 to 4999, the LSB of the instrument is 0.01% or 100ppm. The ADS1115 ADC quotes about the same accuracy numbers:  Gain error, Typical: 0.01%, Offset Error Typical +/- 1LSB. If it does not require major sacrifice, I would make the AFE just a little bit better than the ADC, but not significantly better.


    Doing so will call for a ADC diving grade op-amp, just like the OPA2325, or THS4531, and resistors with controllable match, perhaps with a trim to max out common mode rejection. I do not see a way how CMR can be calibrated in software, other than with good match on the resistors/trimming. Thermal match is out of the question I think, cost prohibitive. However, we can opt for 100ppm/C or so 0.1% precision resistors, at least to have a know thermal behavior. For example: RC0402BR-07100KL, $0.30 each. We might find 0805 too if we search.

Comment

  • Little more on accuracy: If we target 3.5 digit, or displaying from -4999 to 4999, the LSB of the instrument is 0.01% or 100ppm. The ADS1115 ADC quotes about the same accuracy numbers:  Gain error, Typical: 0.01%, Offset Error Typical +/- 1LSB. If it does not require major sacrifice, I would make the AFE just a little bit better than the ADC, but not significantly better.


    Doing so will call for a ADC diving grade op-amp, just like the OPA2325, or THS4531, and resistors with controllable match, perhaps with a trim to max out common mode rejection. I do not see a way how CMR can be calibrated in software, other than with good match on the resistors/trimming. Thermal match is out of the question I think, cost prohibitive. However, we can opt for 100ppm/C or so 0.1% precision resistors, at least to have a know thermal behavior. For example: RC0402BR-07100KL, $0.30 each. We might find 0805 too if we search.

Children
  • in reply to ggabe

    For the prototype for now I used what I consider to be "jellybean" resistors nowadays, i.e. 1% 100ppm thick film (I have these at hand without needing to order anything special), and the board has large 0805 pads, which will allow 0603 and 0805 sized parts to fit equally well, and possibly 0402 at a pinch (I could modify the pads for making that easier).

    For the ADC package, it is a standard 14-pin layout for quad op-amps so it is easy to swap for a better part. The TS9924 has a very sparse datasheet, it's just a low-cost but reasonable part, especially if there's no desire to use beyond a typical room temperature, i.e. just calibrated and used at the approximate temperature in a home. 

    For a nicer part with more defined behavior, perhaps it's best to change the layout for the next revision to use a dual op-amp, since then there's a wider selection. It may push the cost up a bit more, but I think there can still be low-cost options with the dual op-amp.

  • in reply to ggabe

    The smallest OPA2325 package (VSSOP) won't save a lot of space compared to the current quad op-amp package, so I'll have to find a single op-amp that is a lot smaller. Maybe there are easy-to-solder QFN single op-amps, it doesn't need to be special.

  • in reply to shabaz

    (Can't edit).

    SC70 is suitably annoyingly small, so I'll go with this package and pin arrangement (there are several arrangements but this seems most common across more than one manufacturer) for the single op-amp.

    image

  • in reply to shabaz

    (can't edit). i.e., for the next revision, I'll use VSSOP for the dual package, and SC70 for the single package, since I need a total of 3 op-amps. It's all approximately the same cost as the quad-op-amp, so it's still very low cost.