• State Resolved
  • Locked Locked
  • Replies 5 replies
  • Subscribers 49 subscribers
  • Views 722 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.

usable bits of ADC

kevin17291
Prodigy 40 points


1) When using an ADC to sample and convert an AC signal with DC component, do we need to conbine both the DC errors (offset, non-linearity, etc.) and AC errors (SINAD) in order to calculate the usable bits of an ADC?

2) An ADC system used for signal level measurement:  Is there a systematic calculation method to obtain the accuracy of measurment in percentage from the SNR or errors metioned above? Shall we use Vpeak or Vrms for the noise contribution?

  • 0
    TI__Intellectual 1275 points

    Kevin,

    we exclude the DC component when calculating SNR and SFDR for our ADCs, although typically we use AC coupled signals and so the DC offset is only from the ADC and quite small.

    If the DC component from the signal is significant, then it will take away from the full scale range of the converter, which will impact SNR. For a rough estimate you can calculate the loss in the range of the ADC as being

    FS - 0.5*DC

    If you deviate DC very far from 0 offset, like with a DC component within say 20% of full scale + or -, on some ADCs there can also be some decrease (few dB) in SNR due to a noise contribution from the DC component.

    Unless you have a very large DC component, I would just calculate the reduce swing available and use one of the SNR vs. amplitude plots to estimate SNR from the relavent datasheet.

    Robert

  • 0
    TI__Mastermind 29220 points

    Hi,

    I consulted with one of our characterization engineers.

    1) When we calculate SINAD we by necessity take an FFT of a captured buffer of data so that we can subtract out the power in the input frequency bin.  For SINAD, all other frequency bins are counted except bin 0, so nonlinearities are counted.  We do not count bin zero because the data format itself implies an offset of 2**number of bits/2.  This means that we are also throwing away a small amount of noise that is in that frequency bin, but with a large number of bins (say out of a 64K sample depth) then this becomes negligible.  Some noise will alias back into the fundamental bin as well, which will be discarded when we subtract the fundamental bin for summing up the noise power in all other bins, so larger sample depths and thus more fine frequency bins are more accurate.  So SINAD covers nearly everything as much as we can count the sum of all power noise, and effective number of bits is then  (SINAD -1.76)/6.02.

    2) One measure of the accuracy of the measurement is INL in a datasheet, which is usually specified as some small number of lsb's of full scale.  DC offset is also a specified data sheet value.  These specifications would i think dominate concerns of accuracy.  But for considering how SNR itself can influence the accuracy of the measureent, consider that the SNR will indicate how far down the noise floor is from the signal, and that noise is spread out over so many frequuency bins that the noise power in any one bin is very very small compared to the fundamental that it is often 5 or 6 orders of magnitude down, so i think trying to extract noise on top of a signal as a measure of accuracy would pale against accuracy specs already listed in the datasheet.

    I hope this helps, if i have understood the meaning of your questions. 

    Regards,

    Richard P.

  • 0
    TI__Expert 6040 points

    Kevin,

    I only add a little more information for you. In our data sheet (ADS55xx) we do have good definition of SNR, SFDR, SINAD and so on. For your second question, I would add that SNR is signal to noise power ratio and it should use Vrms. If count it in frequency domain, the noise power will be integrated in a certain frequency range, such as from DC (exclude DC and first few of  hamonic) to ½ of sampling frequency. 

    Regards,

    *** Qing 

     

  • 0 in reply to Richard Prentice
    Prodigy 40 points

    Thanks for the reply.

    Baker mentioned in one of his article http://www.analogzone.com/acqt0515.pdf that the total error will be obtained from the square-root-of-the-sum-of-the-squres of TUE and SINAD (the last paragraph above conclusion).

    I think we should consider the integral of the noise contribution across the bandwdith of interest since the FFT will apply to the entire bandwidth even the input signal is a sinewave. When the signal to be digitized is close to the noise level, SINAD should affect the accuracy of the signal measurement. That's my understanding of why TUE and SINAD should be conbined when masuring a signal with both AC and DC components.  Please advise.

    Regards,

    Kevin

  • 0 in reply to kevin17291
    Prodigy 20 points

    Hi,
    Can somebody explain how  did the author in that article convert the TUE 0.1%FSR to 9.97bits(rms)?
    This might be a dumb question, but i would appreciate if someone can help. I'm trying to re-do it and I'm getting a different # from his.
    Also, I'm getting a different value when i added both TUE and SINAD using square root of the sum of the square method.

    Regards,
    Gokul.