Thread Subject:

model fit with discrete Fourier transform?

I have a set of data (conductivity vs length) where there exists a domain of one range of conductivities and after passing certain length we enter a domain of conductivities of different range (so if you image conductivity plotted against length the shape of the curve would be similar to tanh function plus minus noise and less symmetric/smooth).

I need to fit a model to this data and was told that discrete Fourier transform might be up to it, only that the obtained correlation might be a bit complex but still... Yet my (indeed shallow) knowledge of Fourier transform refers to electric applications, even after doing some background reading I cannot figure out how to do it. Can someone advice me on how to tackle this in practise using Matlab built-in routines?

thank you in advance,
mat

On Mar 15, 6:41 pm, "Mateusz Gos" <webmaste...@wp.pl> wrote:
> I have a set of data (conductivity vs length) where there exists a domain of one range of conductivities and after passing certain length we enter a domain of conductivities of different range (so if you image conductivity plotted against length the shape of the curve would be similar to tanh function plus minus noise and less symmetric/smooth).
>
> I need to fit a model to this data and was told that discrete >Fourier transform might be up to it,

I think you were misinformed.

Have you tried a spline fit?

If you have enough data points, a neural net
might be appropriate for approximating the
curve. However, analyzing the model won't
help in describing/understanding the physics
involved.

Hope this helps,

Greg

On Mar 15, 11:41 pm, "Mateusz Gos" <webmaste...@wp.pl> wrote:
> I have a set of data (conductivity vs length) where there exists a domain of one range of conductivities and after passing certain length we enter a domain of conductivities of different range (so if you image conductivity plotted against length the shape of the curve would be similar to tanh function plus minus noise and less symmetric/smooth).
>
> I need to fit a model to this data and was told that discrete Fourier transform might be up to it, only that the obtained correlation might be a bit complex but still... Yet my (indeed shallow) knowledge of Fourier transform refers to electric applications, even after doing some background reading I cannot figure out how to do it. Can someone advice me on how to tackle this in practise using Matlab built-in routines?
>
> thank you in advance,
> mat

Whenever you use the term 'model fit' you need to start
by choosing what model to fit. Only then can you select
a computational routine to actually do the job.

The FFT is of no use, unless your model somehow depends
on the data being expressed in a transformed domain.

Rune