• State Resolved
  • Locked Locked
  • Replies 7 replies
  • Subscribers 46 subscribers
  • Views 621 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.

ADS1248 ENOB

ka88
Intellectual 275 points
Other Parts Discussed in Thread: ADS1248, ADS1234, ADS1232, ADS1232REF, ADS1248EVM-PDK

Hello,

What is the best way to disregard the noise bits in the output code of the ADC?  

For example, if using the ADS1248 @ AVDD = DVDD = 3.3V, 20SPS and PGA = 1...the ENOB due to Peak-to-Peak is 17.3 bits.  This effectively means the lowest 7 bits are noise, and not useful to our signal.  The output conversion table specifies (Vref/PGA)/(2^23 -1) as the LSB.  Would I change this to (Vref/PGA)/(2^16.3 -1) to account for the ENOB?  Please advise.

  • 0
    TI__Guru* 92715 points
    Kevin,


    The output conversion size stays the same ((VREF/PGA)/(2*23-1) as the LSB). The ENOB peak-to-peak shows how the expected amount of noise there is in the measurement. It looks like you get this from Table 6 from the datasheet.

    Normally, this is done at 0V as an input, but for an example, let's say that we're measuring 0V (or shorting the input), with a 2.048V reference, both of which are completely noiseless (we'll also set PGA=1).

    When we take the measurement, we should get an ADC output code of 0. However, there is inherent noise in the ADC itself. In this case, it's about 4.55uV of noise RMS or about 24.74uV of noise p-p. That's taken from Table 5 of the datasheet with the same conditions.

    Here, 4.55uV is about 19.8 codes of noise. If you took one data, the standard deviation of the noise would give you this many codes. If you took about 1000 readings, your peak to peak noise would be about 101.3 codes. The information in Table 5 is used to get the information in Table 6 as well.

    Using a full scale of 4.096V (from +2.048V to -2.048V) your effective number of bits for rms would be log(4.096/4.55e-6)/log(2)=19.78 bits. For peak to peak, this is log(4.096/24.74e-6)/log(2)=17.34 bits as you mentioned in your post.

    So ENOB is a statistical measure of the noise that you see coming out of the ADC and not a change in the measurement value. Note that the noise of the reference can increase the noise that you see as the input signal gets larger and that is why the reference should be a very low noise input voltage.

    To lower this noise, you could average multiple readings. This noise would drop proportionally to the root of the number of averages.


    Joseph Wu
  • 0 in reply to Joseph Wu
    Intellectual 275 points

    Joseph,

    I appreciate the explanation, but I am mostly concerned with how to deal with this inevitable noise.  You mentioned one technique, averaging multiple samples.  If this is not possible, where we can only take one reading, what is another option?  Would it be to "throw away" the noise bits?  So for the example above, I would disregard the lowest 7 bits and keep the highest 17?  If this technique is used, how is the output conversion completed with just 17 bits?

  • 0 in reply to ka88
    TI__Guru* 92715 points
    Kevin,


    Normally, I'd just take the conversion with all 24 bits and accept that there is a small amount of random error in the measurement. I'd also note that Figures 5-8 show histograms of the noise distribution to be Gaussian.

    Are you concerned about the last digits in the readout being noisy? That's a concern that I've seen in weigh scale measurements. If you wanted to truncate the last few bits, you could do that. You're really exchanging the resolution for the appearance of less noise in the readout.

    As an example, take the top 17 bits and convert using (Vref/PGA)/(2^16-1) as the new LSB.


    Joseph Wu
  • 0 in reply to Joseph Wu
    Intellectual 275 points

    Joseph,

    It is for a weigh-scale application.  We have a 120Ohm quarter bridge configuration with a single active gage, Vexcite = AVDD = DVDD = 3.3V, Internal 2.048V reference....and are having trouble measuring very small amounts of force (less than 1Ohm change on the gage) due to the noise.  Currently we are using the ADS1248, although it looks like the ADS1234 would have been a better selection.  Any advice?  So far all we have been doing is increasing the gain as much as possible and lowering the sample rate.

  • 0 in reply to ka88
    TI__Guru* 92715 points
    Kevin,


    For your noise, is it larger than what you would expect from the datasheet? It's possible that you might need some input filtering to reduce any ambient vibration or other low frequency input noise.

    As for the ADS1232 and ADS1234, I believe that they have slightly less noise than the ADS1248. However, the ADS1248 is capable of similar measurements.

    If are interested in the ADS123x devices, I'd take a look at the ADS1232REF. It's basically an evaluation module/reference design. There's a user guide on the website:

    www.ti.com/.../sbau120b.pdf


    Joseph Wu
  • 0 in reply to Joseph Wu
    Intellectual 275 points
    Joseph,

    I am unable to quantify realiably the amount of noise, so perhaps it is the expected amount of noise. The only filtering currently is a 0.1uF ceramic capacitor across each differential input (ie. one 0.1uF cap connecting AIN0 and AIN1). Do you think adding a common mode RC filtering network would help? I was concerned about putting filter resistors in the measurement path due to the very low level input signals from the bridge (10s - 100s of uV).
  • 0 in reply to ka88
    TI__Guru* 92715 points
    Kevin,


    It depends a bit on the environment of your setup, but some modest amount of RC filtering can help.

    I would start by looking at the ADS1248EVM-PDK user guide. On page 31, there is a schematic of the EVM and you should be able to duplicate the input filtering. The user guide can be found here:

    www.ti.com/.../sbau142b.pdf

    I'm pretty sure that the noise performance can be duplicated (or at least come very close) with this EVM. If you wanted to increase the series input resistance, you could, but I wouldn't go larger than 1k.


    Joseph Wu