I am working with EEG recordings about 1,000,000 data pts long, recorded at 4,000 Hz.

I'm generating PSD estimates for these recordings using the matlab "periodogram" function like so:


and plotted like so:


When I look at my plotted data (frequency vs. power), I see these strange, camel-like humps with maxima at every 10 Hz. The humps kind of resemble the plot of the absolute value of a sine wave. These humps accompany and are quite distinct from what I believe to the the non-artifactual baseline data.

When I run the periodogram function on smaller portions of the data as opposed to the full 1,000,000 point recording, the humps and the baseline appear to merge together.

Does anyone know what these humps might represent?


  • $\begingroup$ Could it be "spectral leakage"? These humps continue to show up after trying welch's method and multi-taper estimation. $\endgroup$ Apr 29 '16 at 19:57
  • $\begingroup$ Are you resampling the data to 400 Hz? I ask because your plot seems to go to 200 Hz only. $\endgroup$
    – MBaz
    Apr 29 '16 at 22:02
  • 1
    $\begingroup$ No, the plot is cropped, we are interested primarily in activity below 200Hz. In any case, with some detective work, I've come to believe the humps to be an experimental artifact, due to periodic in-brain stimulation (2 biphasic pulses spaced by 100 milliseconds every 10 seconds). The 100 millisecond spacing is basically 10Hz, right? Could such stimulation cause this kind of effect? If so is there some approach I could take to mitigate it? $\endgroup$ Apr 29 '16 at 23:03
  • $\begingroup$ What do you mean by "in-brain stimulation" in plain English? It looks to me like your signal may have been modulated by a 10 Hz square wave--guessing that those humps are from the absolute value of a sinc function, which is the fourier transform of a rectangular function. $\endgroup$ Apr 29 '16 at 23:21

These are indeed artefacts from the stimulation

x = zeros(1,10*4000)
x(1) = 1;
x(401) = 1;
xlim([0 200]);

enter image description here

  • 1
    $\begingroup$ Yep it's a typical spectrum from a time domain comb, in Andrew's data with some high frequency attenuation. $\endgroup$ Apr 30 '16 at 5:50
  • $\begingroup$ Thanks! I believe this is the case. I'm probably going to just cut out the regions with the stimulation and do Bartlett's method with the remaining blocks. $\endgroup$ May 2 '16 at 16:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.