Your question is very vague. It is unclear what type of analysis you are trying to do, what your input data is (continuous or epoched), and whether you are asking about the amount of data points included in, for example a FFT, or whether you are talking about the actual sample rate of your EEG data. All of these factors are critical to answering this question, and they should be readily available in the methods section of any published paper that conducts a similar analysis. However, randomly entering parameters is certainly not an acceptable solution.

Hi, Brian, its self evident when I say power of EEG signals over frequency and space, that I am using FFT. Sample rate is always fixed and not changed while collecting data. It is an epoched dataset. I mean changing the length of epochs from say 15 to 18 sec brings drastic change to the Topographic plots. Thanks

A FFT is only one of many methods that can be used to quantify EEG data in the frequency domain, so this is not self-evident from your original question. The same analysis could be performed with a TFA using wavelets, a Hilbert transform, or by band-pass filitering and squaring your data in the time domain.

To try to answer your question with the new information given, I am wondering what is the task you are measuring? Because your data is epoched, I assume your design is event-related. However those are very long epochs to use for an event-related design so I am unsure if this is the case.

Hi, I am not measuring ERP's. It's a finger tapping task, I should see FFT peaks at the movement freq, also expected to see ERD in Beta. For TFA using wavelet what do you think about the baseline? It effects my result.

Thanks for the clarification. I believe from your answer that your design is still event-related, just you are quantifying the events in the frequency domain (TFA) rather than the time domain (ERPs).

Because your epochs are so large, it seems like you are planning on analyzing a series of taps at a time rather than individual taps. Is this correct? Also what is the average time between each finger tap?

Hi, The tapping period is 30 sec and followed by 10 sec rest. I should see ERD in Beta at tapping period? Am I correct? What should be my timeline for Time Frequency plotting? Any suggestion on filtering, Finger tapping was at 1Hz.

I am by no means an expert at this particular paradigm, but from my understanding the activity in the beta band is oscillatory, such that it initially desynchronizes but then synchronizes relevant to baseline after each movement. The attached paper should be able to help you determine the correct filtering and TFA parameters for analysis of alpha and beta frequencies in this paradigm.

In general, I would recommend using the middle 8 second period of each rest (unless you have another baseline) as a baseline and probably dropping the first and last tap from analysis to reduce the possibility of any transient effects on the recorded EEG. Then you can calculate ERD/ERS based on this resting period. This will require running two time frequency analyses, one the blocks of tapping and one on the rest periods, and then calculating task-related power: Log(tapping periods)-Log(rests) at each extracted frequency of the TFA. Note that when you calculate task-related power, the tapping data should still be in matrix form (i.e. frequencies x times) while the rest periods should be averaged over times (i.e. a vector of frequencies). So, you should be subtracting the same baseline value from each time point of each frequency of the tapping period.

I would also note that you might have to break each tapping segment into smaller sub-segments (at least 3) because you will need to extract enough time points in your TFA to quantify beta activity which goes up to approximately 30 Hz.

If you want to quantify FFT peaks at the movement frequency, you can just run an FFT on the tapping periods with no baseline removed, a 2 second FFT with 75% overlap should be fine to capture 1 Hz.

Thank you, let me read the paper. I could see the FFT peaks at the movement frequency clearly, also the coherence peaks. But I am working with the time frequency. Let me try the 8 sec as u suggested.

@brian Can auditory stimulus like a metronome can evoke ERP's? I am using metronome with finger tapping.

Hi Debanjan,

I think the first issue that one has to check is the bandwidth that one wants to analyze. One should select sample rates larger than 3-4 times the highest frequency of interest in the EEG signal. For example, if you are interested only in alpha waves (8-12 Hz), your sample rate is 500 Hz, and your interested in a spectral analysis based on FFT, then I think an epoch of 500 ms would be enough.

@Alexander Cerquera

How you linked the sampling rate with the length of the epoch ? is their any rlationship in between them ? for example i am interested in beta band and sampling frequency of my headset is 500 Hz ,then what must be the length of epoch ?

Thanks in advance

Hi, Muhammad

For example, if your interest is only beta bands, and assuming that you consider beta band in the range of 12-30 Hz, then your shortest period would be 1/30 = 33.33 ms, and the longest one 1/12=83.33 ms. If your sample rate is 500 Hz (samples/second), then you would need (33.33 ms x 500 samples)/1000 ms =~17 samples to detect a short beta period and (83.33 ms x 500 samples)/1000 ms =~42 samples to detect a long beta period. Therefore, in an epoch of 500 ms sampled @500 Hz, you would have ~29 short beta periods and ~11 long beta periods. Does it make sense?

Hi Debanjan,

Yes it is very likely that the metronome will evoke auditory ERPs.

Dear Alexander ,

Thanks a lot for your feedback .I have created dummy events to epoch the eeg data so that i can remove the artifacts by various thresholding methods .The length of my epoch is fixed (500 ms ) and the reason of this epoching of data is to just remove the unwanted signal .My frequency of interest is up to 30 Hz and the sampling frequency of the headset is 500 Hz .I cannot take a long epoch because in that case i will loose more data as compare to the short one .I am confused that its right way or not ? Kindly explain what do you mean by 42 samples to detect long beta and 29 for short beta periods ?

When I talked about beta bands in the range 12-30 Hz, I meant that corresponds to periods of 33.33-83.33 ms. Beta periods close to 33.33 ms are short beta periods, and beta periods close to 83.33 ms are long beta periods.

Regarding the artifacts rejection, I don't have too much experience with that issue. Have you checked the tutorials that EEGLAB has to do it? there are some plug-ins that one can use with EEGLAB, for example, one called ADJUST.

Thanks Alexander for the feedback in detail .I am using low channel eegheadset and adjust is not recommended for that .At alpha the sampling rate of the signal is downsampled at 64 Hz.If i take the range of alpha as 8 to 13 Hz then the duration of short alpha period is 77ms and long alpha period is 125 ms .In this way (125*64/1000=8) which means that 8 samples to detect long alpha period.So in an epoch of 500 ms sampled @64Hz i should have 62.5 samples to detect long alpha period .Is the samples mean sampling frequency over here ?