search

csu-hmc / perturbed-data-paper

A paper on an elaborate gait data set.
https://peerj.com/articles/918/
10 stars 6 forks source link

Double Checking Spectrum Results #104

Closed humphreysb closed 9 years ago

humphreysb commented 9 years ago

@spinningplates and @moorepants asked me to look at the new spectrum results because Input and Output seem to match too well. This is with the expectation that there would be more of low pass filter type effect. Here is what I found:

Looking at the time domain, the data on visual inspection seems to match closely at low frequencies but differ at high frequencies. I do see the phase shift that Jason noted. So the time domain does seem to look like a low pass filter to me.

timedomainzoom

I verified this though by calculating the frequency response function from input to output and it looks like a filter too:

frf

(The reason the phase and magnitude look so "noisy" above about 10hz is because the input signal no longer has spectral content there, so it is trying to calculate down in the noise.)

When you are plotting your spectrums, you are using a linear y axis (per @tvdbogert suggestion to not accentuate the high frequencies). I agree with that suggestion, but one downside is that it is harder to discern the differences in the spectrums. Keep in mind that you are showing up to only 8Hz, so the filter effects are not as apparent in your plots. With log scale: logspec

So using log scaling on the y axis all depends on what you would most like to show.

tvdbogert commented 9 years ago

Excellent work, it all makes sense now! I suggest to do semilogy instead of plot to make the Figure similar to Brad's last one. I was wrong to suggest a linear vertical axis.

moorepants commented 9 years ago

Thanks Brad. One thing I notice is that your time series plot doesn't look like Sandy's. Her measured belt speed is very digital. I'm wondering if we aren't using the same thing?

tvdbogert commented 9 years ago

Sandy's time domain plot is zoomed on a short time interval. That is good because it shows the signals better.

skhnat commented 9 years ago

Yes, I gave him the exact files that I had been using.

humphreysb commented 9 years ago

Yes - I used both Sandy's code (frequency_analysis.m) and her files for everything except the FRF plot.

moorepants commented 9 years ago

Thanks a bunch Brad! You've straightened us out on all this.