I've only very briefly skimmed this code and the other thread. My guess is that you're seeing the effect of zero-padding the FFT when you go back to the time domain.
Let's take a simple example of a 10-point sequence and zero pad to 20 points on the call to fft Going back to the time domain and plotting/comparing
xr = ifft(X,'symmetric');
The result for xr is exactly as expected based on the zero-padding the input on the call to fft(), which is equivalent to
If you were to implement the equation for xr explicitly as a function of X (instead of using ifft() ) and use it to compute xr for n > 19, you sould see the periodic extension of xr, and the period will be N = 20.
So, in your case, see what happens if you don't zero pad on the call to fft.
Also, if you remove the zero padding to the nextpow2 you'll have to verify that the rest of the code is correct if the original signal has an odd number of elements (assuming odd-length inputs are of interest).
If, as seems to be the case, the only objective is to generate the periodic extension of the input, there are easier ways to do this than using fft() and developing your own series summation. OTOH, what you're doing is a good way to develop understanding of the underlying concepts.
p.s. IMO the code in "Fourier model.txt" is short enough that it would be better to just copy/paste it into your question (please use the code formatting (left most incone in the Code section in the toostrip) so it's easier for others to see and work with.
