Findpeaks for powerspectrum ignores maxima at first or last index of power spectrum array.

I am trying to use findpeaks to get teh full width half maximum of band different peaks in a power

spectrum. I have used a dummy signal which is a simple sine wave sinx+ sin7x with some noise, but I can see from my power spectrum that there are two frequencies as expected, but findpeaks will only identify peaks that are not at the very first index.

I have enclosed an example. The annotated peak is the only one identified. The other peak, which is the maximum value as well, is completely neglected so I cannot get the width at half max data from it.

[dpb]

That's just the DC component at the origin. Remove the mean from you signal before computing FFT.

[Star Strider]

The findpeaks function ignores them because it cannot detect them as peaks. Use ordinary indexing to return the first and last values of an array.

[Sedo]

@dpb Can you please explain the rationale behind doing this?

[Sedo]

@Star Strider

Even if I index to only the portion of the power spectrum that includes the peak, find peaks still ignores max value since it is at the beginning. Is there a different way you were thinking of indexing?

[Star Strider]

I would test the first value simply with:

[maxpwr,idx] = max(power(1:2))

to test for an initial ‘half-peak’.

If you only ignore the first value, findpeaks should still isolate a peak:

v = [1 2.1 1 0.5 0.5];
[pks,locs] = findpeaks(v)

returns:

pks =
     2.1

locs =
     2

[David Goodmanson]

Hi Sedo,

what method are you using to calculate the power spectrum?

[dpb]

"@dpb Can you please explain the rationale behind doing this?"

Generally the frequency content riding on top of the DC signal is all that is of interest in spectral analysis--the DC level is immaterial. Removing it first leaves more dynamic range for the calculation of the lower-level signal components.

And, I'd not consider the DC component a peak, anyways, even if did keep it hanging around; I suppose there could be some use for it for some purpose but it escapes me what that would be...

You can always get the DC value to normalize with if that is of any significance for particular use (but rarely is ime).

[Sedo]

@ david goodmanson fft and multiplying the the complex conjugate.

[dpb]

"fft and multiplying the the complex conjugate."

NB: The normalization of 2X the 2-sided spectrum for one-sided should NOT include either the DC or Fmax bins; there's only one of them each and all the energy is already there. If don't account for, the apparent DC component is twice the magnitude should be.

See the example at doc fft; it does account for this.