how much dynamic range of  input signal can be measured by AD6645

if the spurious is not counted(because the frequency of the input signal is known),how much dynamic range can be measured by AD6645.

the frequency of input is 3MHz,sample frequency is 90MHz.

after sample ,follow the process:

downsample to 10MHz;

do 65536 point fft;

question :

whether the input singal which amplitude is -100dBm(less than 1lsb) can be measured or not? why ?   

i have done experiment on one ADC evaluation board(16 bit adc). all the signal from 10dBm to -110dBm can be measured.

  • 0
    •  Analog Employees 
    on Apr 18, 2012 12:57 AM

    Hey Chen,

    Using VisualAnalog with ADIsimADC, I was able to simulate the performance of the scenario you've presented.  I've attached the canvas to replicate my results.

    I assumed that you wanted to sample the AD6645 at 90MSPS, digitally filter noise from 10MHz up, and decimate by 9 to see some improvement through processing gain.

    I apologize for the color inversion.  Below is the undecimated results with an input power of -110dBm, 64k FFT.  The noise floor is -122dB.

    Below is the filtered and decimated results with an input power of -110dBm, 64k FFT.  The noise floor is -132dB.

    There is a likely 10dB improvement through filtering.  Does this help answer your question?

    Decimated_AD6645.zip
  • Hey,Tom,

    thanks for your answer!

    however,why the sine wave which amplitude is -110dBm can be sampled effectly.

    -110dBm is far less than 1 lsb of AD6645.

    how to explain this phenomena?

  • 0
    •  Analog Employees 
    on Apr 18, 2012 7:31 PM

    Hey Chen,

    I agree that this seems a bit like magic. 

    The key to understanding this is recognizing the statistical properties of both your signal and noise.  Between subsequent data conversions, signal codes are correlated with previous ones and noise codes are uncorrelated. 

    For correlated signals, doubling the number of samples increases your RMS signal power by 6.02 dB.

    For uncorrelated signals, doubling the number of samples increases your RMS noise power by 3.01 dB.

    Given this, doubling the number of samples of a spectrum containing both signal and noise will net a signal-to-noise improvement of 6.02 dB - 3.01 dB = 3.01 dB each time.

    To resolve signals smaller than an LSB, we need only to take more and more samples to leverage this behavior until the signal power becomes meaningfully higher than the noise power.

  • hey tom,

    thanks,it's magic~,i see.

    So,  doubling the number of samples  will net a signal-to-noise improvement of  3.01 dB each time.

    fft process can also net the same improvment as accumulator,for example,512 point fft can net an improvement of 27.1dB.

    i wonder whether my understand is right or not?

  • 0
    •  Analog Employees 
    on Apr 19, 2012 9:03 PM

    Hey Chen,

    The first thing I'd say is that whether you look at a problem in the time domain or the frequency domain, it is the same problem.  There is no power added to or subtracted from your signal when switching domains.  So what I said previously applies to both an accumulator and an FFT.  In order to take a deeper FFT, you need more samples.

    And to your specific point, any FFT that is 512 times longer will net a 27.1 dB improvement of SNR.  Try it out in our VisualAnalog software and see!