• State Resolved
  • Locked Locked
  • Replies 4 replies
  • Subscribers 46 subscribers
  • Views 476 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.

ADS8912B: Basic question about AC accuracy

Takumi Suzuki1
Mastermind 8610 points
Part Number: ADS8912B


Hi,

Let me ask you following basic questions about AC accuracy.

Q1) According to the datasheet, SINAD and SFDR specifications are as below.

  SINAD : 102.5dB(typ)

  SFDR  : 125dB(typ)

I do not understand why SFDR is better than SINAD(125dB>102.5dB).

Will you teach me the reason?

Q2) Fig21 (datashhet 16page)

There describes that SNR is 102.8db.

I think that as the noise level is around -150dB, SNR should to be around 150dB... 

Will you teach me why SNR is 102.8dB from Fig21?

Thank you for your kindness.

Best Regards,

  • 0
    TI__Mastermind 47630 points

    Hello Takumi-san

    SINAD is the ratio of RMS signal amplitude to the TOTAL sum of all spectral components.  In Figure 21, you must add all spectral components (most are around -150dB, but many near the fundamental are higher).  Adding all of these together will be much greater than each of the individual spectral component values. 

    If you look at figure 21, the highest spectral component is near the fundamental frequency, which is -125dB.  SFDR is the difference between the fundamental (usually near full scale, or 0dB) and the highest spectral peak.

    Below is a zoomed in version of Figure 21; circled in GREEN is the highest spectral component.

    SNR is similar to SINAD, but does not include the harmonic spectral components or the DC component.

    Please take a look at TI Precision Labs - ADCs, Section 4:  AC Specifications for more information.

    https://training.ti.com/ti-precision-labs-adcs

    Also, the Analog Engineer's Pocket Reference guide has some good information starting on page 126.

    http://www.ti.com/tool/ANALOGPRGBK

    Regards,

    Keith Nicholas
    Precision ADC Applications

  • 0 in reply to Keith Nicholas
    Mastermind 8610 points

    Hi Keith-san,

    Thank you for your reply.

    I am studying the technical contents you introduced..but I do not understand well.

    I can see the floor noise is around -150dB from Fig21.

    On the other hand, as ADS8912B is 18bit ADC, the floor noise should be around -110dB, I suppose. (ideal : 6.02*18+1.76=110.12dB)

    I think "-150dB" is too good characteristics for 18bit ADC..

    Will you tell me what I misunderstanding?

    Thanks so much for your support !

    Best Regards,

     

  • 0 in reply to Takumi Suzuki1
    TI__Mastermind 47630 points

    Hello Takumi-san,

    The ADS8912B has an SNR around 102.5dB. Ideally for an 18b ADC, the SNR would be closer to 110dB, but there is noise in addition to the quantization noise which reduces it from 110dB to 102.5dB.

    When an FFT is calculated, it divides this noise energy up into multiple frequency bins. The number of bins depends on how many samples are recorded in the time domain. The total amount of noise energy that results in 102.5dB SNR is divided into many frequency bins. -150dB in the datasheet is the amount of noise per individual Bin. You would need to sum these individual bins up to get to the total SNR value of 102.5dB.

    For example, the ADS8912B is an 18b converter with an ideal SNR equal to 6.02*18+1.76 = 110dB. However, due to additional noise sources inside the device, the actual SNR is closer to 102.8dB. (See Figure 21 in the datasheet.)

    Let's assume that the full scale input range of this device is 1Vrms. The total noise is then 1Vrms/[10^(102.5/20)]=7.5uVrms. If we measure a 2kHz sine wave, and take 32768 samples, then the FFT will have 32768 frequency bins. The total noise will be divided down for each of these bins. In this case, if the total noise is 7.5uVrms, each bin will have SQRT(7.5uV^2/32768)=41.4nVrms.

    If you calculate the ratio of the noise, 41.4nVrms, to the full scale signal, 1Vrms, you get 20*log(41.4nV/1V)=-148dB, which is very close to the noise floor shown in Figure 21.

     

    Regards,
    Keith

  • 0 in reply to Keith Nicholas
    Mastermind 8610 points

    Hi Keith-san,

    Thank you very much for your kind detailed explanation !!

    I understood well.

    Best Regards,