power spectral density of different sampling rate signals

[numhana]

Hi All,

I tried to measure noise from DAQ-system of NI.

My labview program is very simple.

i) Data acquistion from one of port of DAQCard-6036E

ii) PSD (Power spectral density) FFT of signal

iii) Averaging 10 FFT signals

I used two different sampling rates. 200Hz and 200kHz.

PSD should be independent with sampling rate. 

But the noise level (PSD) of 200Hz and 200kHz is 3 order different. It seems like scaling inversely with sampling rate or bandwidth. Do you know why?

[gwd]

Power spectral density should be independent of sampling rate provided (1) you are comparing only frequencies below the Nyquist limit and (2) the noise is white.

At low frequencies you will see 1/f (non-white) noise. The crossover between white and 1/f noise will depend on the particular system (amplifier + signal source).

So there are at least two reasons why sampling rate might matter.

(And let's not forget interference.)

[numhana]

Thank you for the reply.

I think that there are miscommunication.

It is difficult to describe the results. I did several experiments with different sampling rates.

For example, for 200Hz and 40000 samples, I got 1E-8 flat noise level from 0.005Hz to 100Hz.

But for 200kHz and 40000samples, I got 1E-11 flat noise level from 5Hz to 100kHz.

So there are overlap from 5Hz to 100Hz. But from 5Hz to 100Hz, one result was 1E-8 and the other result was 1E-11 (dependent with sampling rate). It should be the same value (independent with sampling rate). It is weird result.

And the results are consistant. If I used different sampling rate (10 time faster (2kHz)), the results were 1E-9 flat noise level.

So I don't think that this result is from 1/f noise.

[Chris_Cilino]

numhana,

What happens as you increase the number of samples taken at the slower sampling rate?

Chris C

[Neil_S.]

numhana,

Thanks for posting to the NI Forums.  The problem you are seeing is likely related to aliasing.  The DAQCard you mentioned does not have any antialiasing circutry.  Noise at higher frequencies can be aliased into lower frequencies when sampling at a slower rate.  Thus when sampling at 200Hz the noise floor can seems higher than when sampling at a higher rate.  Aliasing would also explain why the noise level is scaling inversely with sampling rate.   To measure the noise of a lower band you must either sample at a higher rate or use some sort of low pass filter to eliminate the higher frequency noise.  Hopefully this answers your question.  Good luck with your application.

Regards,

Neil S.
Applications Engineer
National Instruments

[antoine_cea]

I am also trying to measure noise floor with PSD vi apply on a 6132 DAQ. I am understanding the problem with the aliasing but can you recommend any filter to avoid this problem and to be able to perform correct noise floor measurement.

Thank