• State Resolved
  • Locked Locked
  • Replies 4 replies
  • Subscribers 45 subscribers
  • Views 637 views
  • Users 0 members are here
Support feedback
Options
Options
Similar topics

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.

ADS1231: Accuracy of data

sagar shah54
Expert 2125 points
Part Number: ADS1231

Hello,

I am using ADS1231 to calculate the weight of load cell. I am currently using all 24 bits of the DRDY/Dout. The problem is that I am getting not an accurate reading. I mean it remains constant for 16 bits(MSB) and 8 least significant bits vary a lot. 

I am making speed pin to high(80 samples/s) and averaging 10 samples!

Can you help me understand what should I do to improve the accuracy?

Thanks

Sagar

  • 0
    TI__Guru** 113765 points
    Hi Sagar,

    There are number of missing pieces regarding your system and how you have it set up. What voltage is the excitation voltage you are using? And is the same excitation voltage being used as the reference? What is the sensitivity of the load cell in mV/V? Are you using a prototyping system? What are you using for input filtering?

    If you disconnect your load cell and short the inputs to mid-AVDD supply, you will see the best case scenario of your system. This should match the data shown in the noise tables on page 6 in Table 1 of the ADS1231 datasheet. At 80sps you will see that noise-free bits (same as flicker free) is just under 16 bits. It sounds like you are somewhere in the general vicinity with your measurement. You should see at least 1 bit improvement with your averaging.

    It is quite possible that you are picking up 50/60 Hz. You might find that using 10sps with no averaging is actually better than 80sps with averaging. At 10sps you will get a rejection notch at the power line-cycle frequencies.

    Best regards,
    Bob B
  • 0 in reply to Bob Benjamin
    Expert 2125 points
    Hi Bob,

    Yes I am using for prototyping. Yes I am giving 3.3 v for Excitation and reference. The sensitivity of load cell is 1.0 ± 0.15mV/V. I am using a 0.1 uF cap for filtering.

    I will try shorting input to mid AVDD. I am currently able to have 80sps. Thanks for the suggestion.

    Regards
    Sagar
  • 0 in reply to sagar shah54
    TI__Guru** 113765 points
    Hi Sagar,

    When prototyping, you really have to be careful about any noise pickup. Leads must be short (no looping wires). Noise is a large issue, and let me explain why. You have a load cell that will have full-scale output of 3.3V * 1mV/V which is limited to about 3.3mV that you will attempt to measure. This input voltage and any noise will be amplified by the PGA (gain of 128). So it is very easy to pickup EMI/RFI and power line-cycle noise, especially on the load cell cabling.

    From the ADC side you are using a 3.3V reference which gives a full-scale range (FSR) of about 25.78mV (or +/- 12.89mV) and your load cell output of 3.3mV will only use about 1/8th of the FSR. This alone will radically reduce the number of available measureable counts for repeatability of the load cell measurement.

    With an FSR of 25.78mV, the LSB value (or the value of a single count) is about 1.54nV. 8 bits, or 256 codes, is about 394nV, which is a pretty small voltage in the midst of a prototyping system. Air currents or vibrations can cause that amount of noise and more. According to Table 1 in the ADS1231 datasheet, at 80sps and 3V supply the noise will be 549.6nV. This is the best the converter can do with respect to any internal and quantization noise. Averaging can improve this some, but in the end you need to consider the best case scenario which is simply the best the ADC can do as external noise is often much larger than the conversion noise.

    Although the ADS1231 is able to resolve to the value of 1.54nV, the device has limitation due to the noise. Ideally the 24-bit converter reduces to 23-bit (1/2 of FSR) because the load cell will only output in one direction (0 to 3.3mV) and as I mentioned previously uses only about 1/8th the FSR. So out of the total counts of 16777215 counts only about 1/8th of the total (2097152) is in the measurement range. If we look at the full-scale voltage output of the load cell (3.3mV) and divide by the converter noise (549.6nV) you can calculate the total number of usable or measurable counts, which is about 6000. If you divide the maximum value of the capacity of the load cell by the total stable counts you can then determine the measurement resolution. For example a 1kg load cell can best case achieve measurement resolution of 167mg.

    So it is possible to determine a best case scenario by calculation without actually prototyping. Prototyping usually adds noise and it is much better to build up a PCB as a prototype using a good ground plane with the elimination of as much EMI/RFI noise as possible which includes using shielded load cell cabling and correct cable termination. It is not a trivial task to prototype a load cell system.

    Best regards,
    Bob B
  • 0 in reply to Bob Benjamin
    Expert 2125 points
    Thanks for the information. It makes things clear.

    Regards
    Sagar