fInding the tresholhd of signal noise(really long narrow spikes).

Hi,

i have an input EEG signal , i need to filter the data , which means that i need to remove the noise that is the really long narrow spikes, this is how the signal looks:

as you can see there are almost 12 events (large narrow spikes) - and by using the findpeaks function i get almost 2000 pks, which means that i get also the low maximum points, how can i find only the maximum/minimum points of the large spikes or atleast how do find the treshold of for example the lowest line and then plot what's under it?

thanks in advance:)

Answer 1

[Star Strider]

The findpeaks function has a number of name-value pair agruments that can customise its behaviour to give the desired results. One option here is to use 'MinPeakProminencce', and, 'MinPeakHeight' alone or together.

Use findpeaks twice, once on the data vector to get the positive peaks, and once on the negative of the data vector to get the negative peaks. Make the appropriate corrections to get the actual peak amplitudes of the negative peaks, if necessary.

[Star Strider]

Adapt to your signal:

N = 50;
x = 0:(N-1);
v = randn(1, N);
[ppks,plcs] = findpeaks(v, 'MinPeakProminence', 1.25);
[npks,nlcs] = findpeaks(-v, 'MinPeakProminence', 1.25);

figure
plot(v)
hold on
plot(x(plcs+1), ppks,  '^r')
plot(x(nlcs+1), -npks, 'vr')
hold off

.

[rami zaboura]

how do i find the general correct MinPeakHeight / MinPeakProminence?

since my code needs to be run of different files as well, which means that i can't just find the right value of the minpeakHeight of the current specific data that i was given and plot accordingly.

[Star Strider]

‘how do i find the general correct MinPeakHeight / MinPeakProminence?’

Experiment with them.

There are other options as well, such as rmoutliers (introduced in R2018b) and similar functions that can also find them and remove them, depending on what you want to do.

[rami zaboura]

but the problem is that they test my code on different files as well, which means different input, so if i experiment and find the right values on my current file then maybe the minpeakvalue in a different input will be completely different.. i though about for example finding the average peak and multiplying it by a certain factor then take the product as a treshhold and filter the data, but then again i get to the same problem, this time how do i deicde what is the factor... is there a statistical way of doing this?

[Star Strider]

A statistical test of some sort would seem to be a promising direction.

[rami zaboura]

i have actually tried the rmoutliers function and it's giving some nice results, can it be used in my case? (removing noisy eeg signal)

[Star Strider]

You will have to make that determination, since only you have access to all the signals you will be analysing.