Confidence level of a FFT spectrum

[Johnny_Milton]

Hi all,

I would like to know how is possible in LabVIEW, after evaluating an FFT spectrum, to calculate confidence level needed to check the statistical significance of the peaks in the spectrum itself.

Many thanks for your answer!

[JÞB]

Well, I suppose it's time to channel Douglas Adams.  

Apparently you have taken a time domain record and converted it to the frequency domain.   Let's simply call that resulting record "42" 

Now, knowing nothing else, we are left to find the significance of "42"  this will take Deep Thought!

I would hope you can share a few more details about Life, The Universe an Everything pertaining to your measurement. 

[altenbach]

If you have distinct peaks, the magnitude of all neighboring bins can possibly give you the level of random noise and you can probably apply some statistical significance comparing the peak to its surrounding.

How was the signal processed before the FFT (filtering, windowing, wavelet denoise, etc. Is the signal clean, i.e. no spectral leakage, etc.)

Are you mostly interested in magnitude or do you also want error estimates for the phase?

Do you have an example with some typical data?

[Johnny_Milton]

First of all, many thanks for your quick replies.

I believe the problem is independent of the quantity under consideration. The question is what is the level of reliability of each discernible peak inside an FFT spectum, with respect to its background noise: no specific experimental details are needed, it is a general question from the theoretical point of view of the statistical analysis of a sampled database (and i would like to know if anyone has implemented such analysis in LabVIEW).

More specifically, I am analyzing the signal of a pressure transducer placed on the seabed, for the detection of tidal components (sampling rate: 30 min, no interest for phases but only for the amplitude, raw signal analysed by FFT ). Since the tidal components are many, and some of these are place-dependent while others are universal, I would like to know if there is a criterion that, applied to an FFT spectrum, allows us to say: "ok, this is a significant peak, this is not".

Many thanks again for your answers!