Using an FIR Savitzky Golay filter is simple and should give ok results.

Use it to fit a quadratic model to the data. From that you can extract, the smoothed signal, plus gradient and curvature estimates.

The residual of the fit may also be useful, to work out how well the data fits the model. Simple extensions will allow you to further reduce any noise in the measurements.

Just be aware that the filter(s) will introduce a delay. So after you apply it, just advance the filter output to compensate, assuming you are processing offline.

Thanks very much... Thatt was the reply I was looking for. Then have to use polynomial of order 4? Which window size you recommend me? And I have to introduce some boundary conditions?

Respect to the delay. I think OriginLab do the operation you explain because in the graph I don´t see any delay... Look at the graph attached (i think I lose the real peak maximum)

That looks like the output of a smoother. Use a polynomial order of 2 for a quadratic model. I would also recommend using an odd number of points (I.e. 11 not 10) , so the filter output and measurements line up exactly. Does your software offer derivative and curvature variants?

I have used a polynomial of order 2 and an Odd number (11). I think now the filtered signal has more sense... Look:

I don´t know if it offer derivative or curvature variants (I have find but I haven´t see anything). I think there is a more complex Peak fitting but I haven´t used yet... It is very important? I would like to pass a simple filter. The poor result could be compensate with an higher number of strokes analyzed?

Thanks very much. Very usefull for me

Thanks very much, very interesting article... For now we are going to use raw data but in the future we are going to use this filter...

Best regards,

Gabriel

Interesting question, following.

I wish you good luck.