Strange property of sampling noise of NI 6251 board

[daniele.nicolodi]

Hello. I'm testing a NI 6251 ADC board and I have noticed a strange
property of the power spectral density of the acquired data.

I'm sampling a single differential ADC channel, whose terminal are
simply shorted. It does not matter for my experiment but each terminal
of the differential channel is coupled to the ADC trough a simple RC low
pass filter at about 2 kHz.

Sampling the input voltage at different frequencies and comparing the
power spectral density of the acquired data, I observe that the noise is
white (as I would expect for bit noise), but it is higher than the bit
noise for a 16 bit ADC (as I'll understand if there is a source of
additional noise in the ADC).

What surprises me is that the noise level decreases as the sampling
frequency increases.

I computed the effective number of bit of the ADC, inverting the well
known formula Sv^(1/2) = Delta_V / sqrt(12 * f_s) where Delta_V is the
ADC resolution and f_s is the sampling frequency. I obtain that this ADC
has an effective number of bits of 13.5, at all the sampling frequencies
I have tested so far (between 1 and 100 kHz).

Is someone able to explain me why the ADC noise has this behavior?

Thanks. Cheers,
-- 
Daniele

[reikira]

Hi Daniele,

I think you can find your answear in the following documents:

The Differences Between Accuracy and Bits of Resolution

How Do I Calculate Absolute Accuracy or System Accuracy?

Serena

[daniele.nicolodi]

Hi Serena, thanks for your answer.

I'm well aware of the fact that there are multiple noise sources in and ADC system, and that I should not considere just the digitization noise. However, what puzzles me is the fact that my ADC board behaves exactly as an ADC board with 13.5 equivalent bits. I would expect aditive noise, coming from other components of the system, to not be white and to not not scale with the sampling frequency as the digitization noise.

Cheers,

D.

[ChrisR.]

Daniele:

What you're seeing is expected.  The whiteness of the additional noise sources is irrelevant.  The issue is that the bandwidth of the board does not change when you change the sample rate.  So you see all the noise all the time.  If you reduce the sample rate, that noise is distributed over a smaller range of apparent frequencies (some of it is aliased), so the noise density goes up.

Chris 

[daniele.nicolodi]

Thank you Chris. What you write makes perfectly sense.

I was put out of track by the fact that the noise power spectral density scales exactly as bit noise of a strange 13.5 effective bit ADC... What I am still wondering is: why then, in all the documents I read, NI always gives noise figures regardless of the sampling frequency?

[ChrisR.]

Can you give examples?  On DAQ products we specify noise independent of sample rate because it is independent of sample rate.  The noise density is not, but we don't specify noise density; we specify total noise.

Chris 

[daniele.nicolodi]

Sorry. I think the issue is impedance missmatch between engineers and physicists gergo. Thanks for your help.