>have tried wave ports with integration lines, and it gives me the same curve, but without spikes.
100%, the spikes are higher order modes crossing the fundamental mode. With the field vector you lock the mode and filter them out.
So, the results should be correct for your model and setup.
For better understanding, you should make an eigenmode section of your waveguide, and sweep the phase in order to see the dispersion, which will give you a hint on impedance. If it is bent somewhere, that will be the answer.

Don't forget that when no boundaries of the model are defined for "airbox" in HFSS, PEC condition is used. You should define some radiation and move the boundaries far away.