This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

ADS1261-Q1: Combining DC errors and noise

Part Number: ADS1261-Q1

Hi again,

Since the last thread, I've opted for a differential measurement approach to minimize error as much as I can (see diagram below). With this new approach, I'm trying to determine the voltage/temperature resolution the system is capable of measuring. I'm getting confused on how to go about accounting for both DC error sources and noise. Here are the sources of DC error that I've identified in the system. 

  1. RefOut
    • Initial tolerance
    • Drift/droop over temperature
    • Load regulation
  2. ADC input current
    • Typical differential current
    • Differential drift over temperature
  3. Beginning of life resistor tolerances
    • TCC sensor
    • Bias resistor
    • Low pass filter series resistor
  4. Resistor changes over temperature
    • Bias resistor
    • Low pass filter series resistor
  5. Mismatched voltage drops
    • PCB traces
    • Low pass filter series resistor
  6. PGA gain error 

Based on our previous discussion, I will exclude the RefOut error due to the ratiometric properties of the measurement. Using the remaining RSS'd error sources, I can calculate a worst case range for the thermistor voltage reading across the ambient temperature range. Here, worst case indicates the narrowest range as that would reduce the resolution I'm capable of measuring. Using the datasheet values for temperature drift/coefficients I can approximate a voltage vs temperature curve as well.

In parallel, I've also calculated the noise-free resolution that I've translated into an LSB size. Where I'm getting confused is how I can incorporate the two separate calculations together to arrive at a realistic error-free voltage measurement resolution. 

Thanks,
Drew

  • Hi Drew Nakamura,

    You should also consider the offset and linearity of the ADC, though these will probably be pretty small. I would expect that the gain errors of the ADC (including the PGA) and the bias resistor will be the largest sources of error (and their respective temperature drifts, unless your system temp range is small)

    The noise acts as a floor with respect to your error. For example, if your DC error was 1uV but your noise is 1mV, you could not resolve a 1uV signal so that accuracy specification is meaningless.

    Typically with high precision ADCs it is the other way around: the noise is less than the calculated DC error.

    Also, if you try to calibrate out any DC error, the calibrated error can only be reduced to the level of the noise.

    -Bryan

  • Hi Bryan,

    Thanks for the response! That all makes sense.

    -Drew

  • Hi Bryan,

    Follow up question. Which error sources will be corrected for during a DC error calibration (e.g. sweeping the board over the ambient temperature range and recording the voltage/temperature curve)? 

    Assuming none of the temperature relationships mentioned above have random variation I'd expect everything to be calibrated out. At the very least I'd expect the beginning of life resistor values and nominal ADC input current to be calibrated out. 

    My calculations are confirming your expectations that the bias resistor causes the most error, so if I can convince myself that those errors are repeatable and can be calibrated out then that would significantly improve the measurement accuracy.

    Thanks,

    Drew

  • Hi Drew Nakamura,

    Are you going to calibrate your sensor as well? If not, how are you going to bias the system to perform the calibration?

    Assuming the source input is ideal i.e. 100% accurate, it should be possible to calibrate out most of the signal chain error (to the level of the noise). There will be some residual linearity error that cannot be removed, unless you run a linearity sweep. However, this will be expensive in terms of actual cost as well as time. Especially if you are also calibrating over temperature.

    Also, if your temperature drift error is nonlinear, this will not be removed by the calibration unless you calibrate at a very small temperature increment.

    Realistically, it makes sense to identify all of your sources of error and then determine which is the most critical to remove. A simple offset and gain calibration might be sufficient to get to the required accuracy level. In other words, don't create a super intense calibration routine if it is not necessary.

    -Bryan

  • Hi Bryan,

    We don't have plans to calibrate the sensor itself outside of the system. It'll be part of a larger TTV package that will arrive installed on the board. The plan is to power only the ADCs and microcontroller and not power any of the heating elements inside the TTV during calibration. That on top of at least an offset calibration would be the extent of the calibration routine. It sounds like it would require too many assumptions to convince myself that the errors will all cancel out during calibration so I will leave them in my analysis for now.

    Thanks,
    Drew

  • Hi Drew Nakamura,

    If you are performing a simple offset calibration with the inputs to your system shorted, then this will remove the offset and nothing else. So yes, you will need to consider all other error sources in your analysis

    -Bryan

  • Thanks Bryan. My last question on this issue is related to reducing the impact of Rbias error from the measurement. Regardless of if I use the RefOut voltage or IDAC current to bias the thermistor it seems that this error source cannot be removed. I can minimize it by looking into expensive, thermally stable resistors but my calculations show that Rbias error can add up to 1.2C in error. I'd like to be able to measure changes as small as 0.1C, but without assuming this to be an ideal resistor I don't see how that's possible.

    -Drew

  • Hi Drew Nakamura,

    There are generally three things you can do if your system performance does not meet your accuracy requirement:

    1. Relax your accuracy requirements
    2. Find the largest source of error and replace it with better/higher accuracy components
    3. Calibrate

    It sounds like you cannot do #1 and #2, which leaves #3 as your only option. Let me know if I can offer any more guidance about calibration

    -Bryan

  • Thanks Bryan. I will see how far the first two items will take me for now. If I have calibration questions in the future I will create a new thread. Thanks again for all your help.

  • Hi Drew Nakamura,

    Sounds good, I am glad we could help!

    -Bryan