F0 coding of speech-like stimuli using the OpenBCI EEG recording

cananthk

I am trying to investigate the F0 coding of speech-like stimuli using the OpenBCI EEG recording. The F0 range is between 100 to 126 Hz and the formants of the speech are between 250 to 4000 Hz. The minimum sampling frequency of the stimulus is 10 kHz.
Can we accurately estimate F0 for this experimental paradigm using the OpenBCI EEG recording? Also,
If I use 8 channels with Fs = 250 Hz, only F0 information will be extracted but do you think the formants will be aliased and might distort the EEG signals? Please let me know your input.

wjcroft

Hi Anantha,

If your signal is in the range of 100 to 126 Hz, then a 250 Hz sample rate is not going to do well at all. 'Technically' the Nyquist limit says a 250 Hz sample rate can record a 125 Hz signal, but that is with extreme amounts of aliasing. For better signal quality, instead of only 2 samples per waveform, 5 samples per waveform would be better. Three is not much better than two. Four still has considerable potential for aliasing.

250 / 5 = 50 Hz max input signal frequency for sampling at 250 Hz.

William

cananthk

Hi William,
Thanks for your response. As the male fundamental frequency (F0) lies between 80 - 175 Hz, I have to use the F0 between 100 - 126 Hz.
Here are the three options:
1) With a 250 Hz sampling rate, I can estimate the aliased frequencies (or fold-over frequencies) beforehand. After acquiring the EEG signals (mainly gamma waves), I can apply a post-filter to minimize the effect of aliasing.
2) Can I record the EEG signals with a higher sampling rate? If yes, then please specify.
3) By looking at these constraints, I can alter the F0 between 80 - 100 Hz, but be in the male F0 range.
This would perhaps minimize the aliasing effects in actual recording. With the help of post-filtering, I can further minimize the aliasing effect.
Let me know your thoughts.
Ananth

wjcroft

@cananthk said:
...
1) With a 250 Hz sampling rate, I can estimate the aliased frequencies (or fold-over frequencies) beforehand. After acquiring the EEG signals (mainly gamma waves), I can apply a post-filter to minimize the effect of aliasing.

Imagine a 125 Hz input frequency being sampled at 250 Hz. If the samples happen to occur at the input peaks, your signal looks like a triangle wave. As you know triangle waves have a TON of odd frequency components in the FFT.

https://en.wikipedia.org/wiki/Triangle_wave

Now if the samples occur at the null points, then the result is equally disasterous. Mid range samples will look like triangle waves of lower amplitude.

2) Can I record the EEG signals with a higher sampling rate? If yes, then please specify.

Not currently.

3) By looking at these constraints, I can alter the F0 between 80 - 100 Hz, but be in the male F0 range.
This would perhaps minimize the aliasing effects in actual recording. With the help of post-filtering, I can further minimize the aliasing effect.

Refer back to the previous comment suggesting that FIVE samples per waveform is more realistic.

---

A consideration we forgot to mention previously, is that skull conduction of gamma and above is extremely weak. Thus even at 40 Hz, gamma amplitudes are at most a few microvolts. At such levels, noise frequently overwhelms the actual signal. I'm afraid that at 100 Hz you are not going to find anything.

EEG recordings at higher gamma range do occur, but only during neurosurgery when the electrode is placed directly on the cortex. This is called cEEG or ECoG, cortical EEG.

https://www.google.com/search?q=cortical+eeg

Billh

cananthk,
I am confused about what you want to measure.

Isn't your male voice "F0" an auditory frequency? If so, are you assuming that you will be synchronizing the EEG frequency with your auditory frequency in some way? I do not think such synchrony occurs beyond the the inner ear, and you are planning on doing EEG, not EcochG, right?