Better explanation on how to choose the good design matrix when scaling FIR coefficients

[benoitvalery]

Hi, I'm trying to expand the GLM FIR Analysis tutorial to a more complex case than the one presented. There is something which is not clear to me in the Plot the response from a single condition section: I wonder how and why fir coefficients are scaled to an individual design matrix. From what I understand, each individual record is estimated via a particular design matrix, which depends on the individual signal length -- so that I do not understand why this kind of design matrix (the dm variable in the tutorial) is used to scale FIR coefficients (df_sum), which is not an individual output.

Could the tutorial elaborate a bit more about the way to select the appropriate design matrix to scale the fir coefficients ?

If needed, I can elaborate more my question.

Thanks a lot for your great work.

[rob-luke]

Hi @benoitvalery, Sorry for taking so long to respond to your message, somehow I didn't see it. In the meantime, have you found a paper that discusses appropriate scaling details? I would be pleased to reproduce any articles in MNE-NIRS to demonstrate how to achieve best practices from the literature.