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.

REF1933AIDDCT 3.3v as reference for ADS1256

Other Parts Discussed in Thread: ADS1256, REF1925, REF1933, REF5025

Hi,

I'm connecting the ADS1256 using REF1933AIDDCT as the reference (VREFP) at 3.3v and AINCOM at 1.65v. VREFN is connected to AGND. AVDD is 5v (feeding from RaspberryPi 5v pin) all ground is connected eventually to RaspberryPi ground pins.

Settings for ADS1256 are PGA=2, SPS=2000. I use only 3 single ended channels, cycling through them as described on page 21 (fig 19) in the manual.

Status register is ACAL=Enabled, Bufen=Disabled. 

1. Is there a way to estimate the ENOB for that configuration?

Cheers,

Ran

  • Hi Ran,

    Welcome to the TI E2E Forums!

    Please note that the ADS1256 can only take a maximum 2.6V reference, so the REF1933 will not work with this ADC, but you could use the REF1925. That said, I'll answer your question using the REF1933 as an example...


    It is difficult to say what the noise performance of your "system" will be when it's all put together. The ADC ought to perform as specified in the datasheet; however, overall system noise performance can degrade due to noisy supplies, poor layout, noisy input signal (from sensor or signal conditioning) and reference noise.

    When the ADC inputs are shorted, you ought to achieve noise performance similar to the datasheet specifications. If not, then the issue is most likely due to a noisy power supply or poor layout.

    • Note: If you are jumper wiring to the Raspberry PI, then you could be introducing a lot of noise form these connections. Jumper wires introduce a lot of loop area (as opposed to PCB traces routed directly above a ground plane), so connections tend to be very inductive and sensitive to RFI.

    Assuming that your power supply and layout are clean, the input signal and reference noise will be the other things to consider...

    • If you apply a noisy input signal, the ADC's digital filter will only filter out so much of the signal's noise; otherwise, you're stuck measuring a noisy signal. When you take measurements with a signal applied, try to use a low-noise signal source. Sometimes I've seen people try to use a AA battery for this purpose; however, with a 24-bits ADC you'll certainly be able to observe the battery's noise.

    • The reference noise will scale with your input signal. So if you notice that noise gets significantly worse as your input signal gets larger, then you know it is most likely related to the reference (this is because the ADC measures the ratio of VIN/VREF. When VIN = 0 V,  the ratio is 0 and reference noise will have no effect.) You can estimate the ENOB due to the reference noise...refer to my blog post on this very topic:
      So for example: at 2000 SPS, the digital filter bandwidth of the ADS1256 is just under 1 kHz. The REF1933 contributes about 27 uVrms of noise with this bandwidth (after summing the 1/f noise and broadband noise contributions - In contrast the ADS1256 only contributes 2.4 uVrms for 2kSPS and PGA = 2 V/V). Therefore, the ENOB of the ADS1256 + REF1933 with a full-scale signal applied would be approximately: log2[FSR / (ADC + REF Noise)] = log2[4*3.3V / 2V/V / SQRT(27uV^2 + 2.4uV^2)] = 17.9 bits.
       
      Using the REF1925 as a more realistic example, the reference noise is only 75% (2.5V/3.3V) of the REF1933 and the FSR is also 75%, so the ENOB ratio is still about 17.9 bits.

    ...If you use additional sensor conditioning, you would also need to estimate the noise contribution of any opamps and multiply it by the input gain. (Also keep in mind that with the ADS1256's buffer disabled, your input signal needs to be driven (low-output impedance) to avoid additional gain error.


    Best Regards,
    Chris

  • Hi Chris,

    Thank you for the very good reply. 

    You wrote: "Please note that the ADS1256 can only take a maximum 2.6V reference, so the REF1933 will not work with this ADC"

    I probably missed that part of the specification and I have a PCB with the 3.3v as a reference already running, giving a more-or-less the expected values from my sensor (LIS344AHL). Does it mean the ADC is only using 2.6v out of the 3.3v? or that the performances will be unpredictable? It is working for over a week now. Looking at the specification book, page 17, it reads:

    "To keep these diodes from turning on, make sure the voltages on the reference pins do not go below AGND by more than 100mV, and likewise do not exceed AVDD by 100mV".

    Since I use AVDD of 5v, I thought a ref of 3.3v would be just fine. I wonder if the input 2.6v limit applies only if using the buffer?

    The sensor I use has an output range of ~ 0.3v-3v, with zero-offset at 1.65v. using the REF1933 seems like the best choice to me. Can you recommend a better solution in light of the ADS1256 2.6v ref limit? 

    Cheers,

    Ran

  • Hi Ran,

    I don't know exactly what the issue is with using a reference voltage greater than 2.6 V. It's possible that because your not too far outside of the specification that you don't see an issue now, but you never know if it can become a problem later (for example, if you test the device over temperature).

    You would okay with switching the REF1933 for the REF1925. The ADS1256 is unique from most ADCs, in that it can measure voltages of +/- 2*Vref. Therefore, with a 2.5V reference you would still be able to measure your 3.3V sensor signal.

    ...Otherwise, I would recommend using a lower-noise reference, such as the REF5025, to preserve the noise performance of the ADC and achieve higher resolution (The REF5025 will contribute about 2.8 uVrms of noise under similar conditions, which is much more comparable to the ADS1256's noise performance).

    Best Regards,
    Chris

  • Hi Chris,
    Thanks again. One last question:
    How did you calculate the noise contribution of 2.8uVrms of the REF5025 (and also for the REF1933)? I couldn't find those values in the datasheets.

    Cheers,
    Ran
  • Hi Ran,

    No problem! I got the noise figures of the REF1933 from the electrical characteristics table:

    (FYI: For some reason the current REF50xx datasheet is missing these specifications - I used previous known values of "7.5uVpp" for the 1/f noise and "5nV / √(Hz)" for the broadband noise)

    You can then use these values to estimate the total noise in the same way that you would figure out the noise from an opamp. For example, the broadband noise is calculated as the noise spectral density times the square root of the bandwidth: 3.3V * 0.25ppm/sqrt(Hz) * sqrt(1000 Hz) = 26 uVrms.

    Here is a video that explains this calculation in more detail:

    Best Regards,
    Chris

  • Hi Chris,
    That is very very useful, Thanks!
    However, looking at the video (minute 8 etc.), shouldn't I also multiply Hz by 1.57 to convert the spectral density to RMS? I assume you used 1000Hz since the sampling rate is 2K SPS?
    Cheers,
    Ran
  • Hi Ran,

    The noise spectral density is almost always given in Vrms/sqrt(Hz), so I don't think you need to do any conversions to get to the RMS noise value. In the case of the REF19xx, the noise spec is normalized to a PPM value so there is one extra step to calculate the voltage noise density.

    The other important parameter is the noise bandwidth. For my calculation I used the approximate noise bandwidth of the ADC's digital filter at 2kSPS. This bandwidth is usually smaller than any RC filter you put in front of the ADC. There is an Excel calculator here that can help you determine this bandwidth:

    For a SINC filter, the transition band is fairly steep; therefore, the noise bandwidth is fairly close to the -3dB bandwidth. However for an RC filter, the transition band roll-off is not as step and the "1.57" factor is needed to estimate the noise bandwidth.

    Best Regards,
    Chris

  • Hi Chris,

    Thanks for the file. very helpful. In the Noise table in the file It reads 1760 for 2k SPS. so I should use that instead of 1000Hz you used in your original example? i.e. instead of

    3.3V * 0.25ppm/sqrt(Hz) * sqrt(1000 Hz) = 26 uVrms

    it should be:

    3.3V * 0.25ppm/sqrt(Hz) * sqrt(1760 Hz) = 34.6 uVrms?

    Regrading the RMS etc.: I try to plot the expected noise from the ADC on a power spectral density plot to compare with NHNM of seismic signals (Peterson, 1995):  Peterson_1993_bg_noise_model.pdf

    I convert the Vrms to m/s^2 (where the sensor output is 0.66 v/g). However, since its RMS (same as standard deviation?) it will require to multiply by 2.506 (following Peterson, 1995) or by 6 if we assume a 3sigma interval and a 97% confidence in order to convert to peak-to-peak spectral density and compare the noise directly to the PPSD noise?

    Cheers,

    Ran

  • Hi Ran,

    Opps...those values are wrong. Use the "digital filter" tab result, shown here:

    There was an error in my initial effective noise bandwidth calculation. I corrected it on the "digital filter" tab, but forgot to update the "noise table" page. Sorry about that!

    Yes, Vrms is essentially the same as the standard deviation (sometimes Vrms includes the average DC voltage of the signal, but in this case we're treating noise as an AC signal with zero mean; therefore Vrms equals the standard deviation). To convert the RMS voltage to a Peak-to-Peak voltage, you would then multiple by a crest factor (typically 6 or 6.6) depending on the probability confidence you're after.

    Best Regards,
    Chris

  • OK. Thanks. That is very helpful. I guess now I need to figure out how to adjust the reference voltage. Since the sensor's output is only 0.3v-3v the most effective would be a 1.65v reference that will be used also as a AINCOM and PGA of 1 to get a full scale +-3.3v. However, I'm not sure I could find a 1.65v low-noise reference.
    Cheers!
    Ran
  • Hi Ran,

    I wouldn't worry too much about trying to get a 1.65V reference. When your working with lower resolution ADC, picking the optimal reference voltage can be beneficial to keep the quantization noise low (quantization noise is proportional to the LSB size, and therefore the reference voltage as well). However, at 24-bits, the thermal noise is much larger than the quantization noise. I usually just recommend using the largest reference voltage in this case to get the widest input range. Then if your input signal is over-ranged, you have more headroom to capture it.

    Best Regards,
    Chris