I am very new to analyzing EEGs and have a few questions about Fast Fourier Transformations (FFTs) and Power Spectral Densities (PSDs).
I am conducting research that is looking at the effects of cell phones on brain waves. I want to see if the presence of a cell phone in talk mode affected the brain waves when compared to the baseline EEG.
Here are my questions:
1. If I run an FFT on the raw baseline EEG and the raw exposed to cell phone EEG, would I just be able to look at the x-axis to see the delta, theta, alpha, and beta waves? Meaning, could I just look at the plotted points between .5Hz and 4Hz on the x-axis to know the delta wave FFT values? Could I also do this when looking at the PSD graph?
2. In what instances would it be better to look just at the FFT values and in which instances would it be better to look at the FFT values?
3. The EEGs I ran are on BioPac pro software and when running either the FFT or PSD I can either manually break apart each EEG into epochs and run either the FFT or PSD or I can run it on the entirety of the EEG. What are the benetfits of breaking an EEG into epochs to run either an FFT or PSD?4. When runnning the EEGs I had IIR acquisition filters on to separate delta, theta, alpha, and beta waves into different channels. What difference would it make if I separately ran either an FFT or PSD on each of those channels than if i ran it on the raw EEG and then separated out the different waves
Hi Laurie.
I'm new to the field too, but come from an electrical engineering background.
I strongly, strongly suspect that you will run into problems collecting any data from your experiment at all due to the following three reasons:
- All wires act as antennas (and inductors, too). This includes EEG leads. They will pick up any nearby alternating currents, including signals from mobile phones...
- EEG signals are naturally very weak, partly due to the low power per unit surface area of the brain, and partly because it's low pass filtered to hell by the skull.
- The inverse square law dictates that the voltage generated across an antenna by a point source (such as a phone) is proportional to 1/(r^2), where r is the distance from your source. This means that the phone signal will be very, very strong - far stronger than power line noise which can already dominate EEG signals.
I'd suspect that even with pre-amplifiers, you'd be looking at a signal-to-noise ratio of far below 1. Of course, I could be wrong. Definitely worth taking the 2 minutes to put an electrode on then make a phone call next to the wire. I suspect all you'll see is artifacts caused by the phone, though.
Regarding the first question, raw data is a plot of voltage vs. time. There is no frequency axis until you've taken your FFT. That's why the lines are so squiggly.
Regarding the last question, I've actually run an FFT on 8 seconds (!!!) of data before and wondered why it didn't work. The simplest way to explain it is that FFT trades time-domain resolution for frequency-domain resolution. If you run your FFT on large time bins, you won't be able to see how your signal changes over time. To see how different frequency components change over time, the best way I know of is through a spectrogram.
If you calculate the FFT of the whole signal your spectral estimation has a variance that is the same as the value estimated, that is, the uncertainty associated to the value estimated is correct is high. As you divide the record into M smaller segments, the variance reduces with sqrt(M). The technique of using shorter segments and averaging the spectra was first proposed by Welch in 1967 and any book on DSP that is worth its salt mentions and explains it.
FFTs and PSDs for EEGs. Available from: https://www.researchgate.net/post/FFTs_and_PSDs_for_EEGs [accessed May 26, 2017].
- Badges
- Science topic
- Similar topics
- Geoscience
- Cartography
- Mapping