ADS1672 noise floor

Mr Hillis -

The text that you describe is the method that is used to test the performance...ie short the inputs and observe the test plug-ins.  Typically for input short, the histogram is the best test plug-in to use to observe the results. If you want to use the FFT, you may want to change the Y-axis to dBFS, otherwise you may not see what looks like valid data with the default scale of dBc.

Greg,

I am looking at dBFS, not dBc. I tried inserting a screen capture of the ADCPro window but it did not take. I will try again.

I am looking at the spectral plot because I am also interested in any artifacts (idle tones etc.).

Craig

Looking at the plot below, the signal power at 36kHz is -96dB relative to full scale. The SNR is -21dB which would put the noise power at -75dB relative to full scale. According to the data sheet, at fDATA = 625kSPS the minumum dynamic range is 105dB.

I have used 14 DC Leakage Bins in the FFT tab to make sure the DC bin is not affecting the reading.

Is it possible with the evaluation board to reproduce the same frequency spectrum plots as seen in the data sheet? If so how? The spectral noise I see is much larger than advertized in the data sheet.

Mr Hillis -

You probably want to make sure that you are using coherent sampling.  There should be a calculator in the ADCPro that will help with determining the right frequency, number of points, etc. to do this properly.

If I am not injecting a tone do I need to be concerned about coherent sampling?

If I am not coherent sampling would the noise floor of my plots be 20dB greater than the noise floor in the data sheet. The plots in the data sheet were created using a 65536 point FFT. I have used larger FFTs and still have higher noise floor. If you are not doing a per root hertz correction on the plot, the noise floor should drop by 3 dB for every doubling of the number of points.

Mr Hillis -

Upon further investigation, it appears their may be some incorrect scaling for the display when the inputs are shorted.  The data viewed in the time domain looks to be correct when we calculate out the numbers.  We will need to investigate why the plot shows incorrect values with no input value versus a small signal.

Greg,

Thanks for the response. I trust this means that the problem is either in the display software or the DSP on the mother board, and not in the ADS1672 chip.

Craig

Mr Hilllis -

we looked at the data for the input short condition in the scope and it calculates out correctly, so it appears there is some kind of display offset that is behaving incorrectly.  This is not a function of the actual ADC.