• State Resolved
  • Locked Locked
  • Replies 3 replies
  • Subscribers 45 subscribers
  • Views 510 views
  • Users 0 members are here
Support feedback
Options
Options
Related

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.

Total Error for AD1115 using +/- 256mV FS setting

Mark Tucker95694
Intellectual 340 points
Other Parts Discussed in Thread: ADS1115, ADS1118

I'd like to use the ADS1115 for a thermocouple sensing application.

Since the TC signals are so small, I intend to use the +/- 256mV Full Scale gain setting (and differential inputs per SBAA189).

I'm trying to find the total error versus input level for that particular gain setting. Figure 19 in the data sheet shows the total error for +/- 2.048V full scale. Can I just take that plot, scale horizontal and vertical axes by a factor of (0.256/2.048) and get valid results?

FWIW, assuming this is a legit approach, a TC signal of only a few millivolts p-p will have a max error of ~ 25uV. This sounds about right and corresponds to 3 LSBs. Is this scaling approach copacetic? 

If not, can you provide a similar graph for +/-0.256v FS? 

Thanks,

Mark

  • 0
    TI__Guru** 113765 points

    Hi Mark,

    Unfortunately the device doesn't scale well with respect to error. I don't have a graph for the conditions you want to use.  The total error is just the RSS of all the error sources.  In the plot you mentioned you will notice that gain error dominates.  You can see the effect of gain error with respect to temperature and full-scale range in Figure 6 of the datasheet.

    Bottom line, it is much worse that 25uV at full-scale of +/- 256mV.  In an uncalibrated system, with a much lower than full-scale input, the error becomes much less significant.  You can see the effect if you draw two lines that both start at zero, but have slightly different slopes.  Close to 0 the lines don't diverge that much, but as you move farther away the divergence is greater.  If the voltages you are measuring are much smaller than full-scale, you will not see as dramatic effect compared to the full-scale measurement error.

    All is not gloom and doom however as the gain error can be calibrated out.  You can recalculate the slope by using a two point method.  One way of doing a two point calibration is to use 0 deg C and 100 deg C.  From these two points you can recalculate the slope to use as a correction factor.

    One thing to keep aware of is cold junction compensation (CJC).  If the cold junction has uneven temperature or if there is a large degree of error in the CJC measurement you result will also have error.  You might want to take a look at the ADS1118 and the associated EVM.  This device is very similar to the ADS1115 and includes a precision temperature sensor that can be used for the CJC measurement.  The biggest differences between the parts are the communication interface and the alert threshold comparator.  The ADS1118 is SPI compared to the ADS1115 I2C interface, and the ADS1118 does not have the comparator for alert.

    Best regards,

    Bob B

  • 0 in reply to Bob Benjamin
    Intellectual 340 points

    My TC inputs will be on the order of 2-4 mV pp. That should certainly keep my operation at or near the intersection of the error limits in the center of the graph - correct?

    For the +/- 2.048V scale shown in figure 19, the minimum error is roughly +/- 200uV at the center of the graph. I scaled that by (256/2048) to get the 25uV value.

    So that scaling is overly optimistic - even at these low signal levels? 

    Thanks,

    Mark

  • 0 in reply to Mark Tucker95694
    TI__Guru** 113765 points

    Hi Mark,

    The dominant error at the center of the graph is usually offset (but it could be linearity.)  Offset error doesn't scale (nor does linearity), but this too can be calibrated out.  The total offset error will depend on supply voltage used and whether you make the measurement differentially or single ended. 

    If you short the inputs, you can subtract (or add if the offset is negative) the correction value to make the result 0.  This same factor would be used to adjust the subsequent readings.

    Best regards,

    Bob B