@Joe Riel Thank you! It worked very well in terms of speed (many times faster than the other computation). Here is the result:

The only thing you have to take care is the maximum integration step (maxstep=300e-9 in the former example). In the following example I used the default step for rkf45 and it clearly didin't work.

@dharr Thank you very much! I left it running for some hours and got the graph done.

Yes, the discrete approach is what is done in practice and I'm trying to graphically compare both approaches for a class I'm preparing.

Sliding_AVG_01.mw