HFSS inaccuracy - "rectangular waveguide modal excitati

I want to simulate a large (um-scale) system in HFSS using low-wavelength (infrared) radiation.

But when the frequency is sufficiently high, HFSS gives erroneous results ? the S-parameters all of a sudden oscillate randomly between 0 and 1. This is apparently due to a "rectangular waveguide modal excitation", whatever that is (see the file I posted at https://www.edaboard.com/viewtopic.php?t=91651).

I can "solve" the problem by making the system and thus the waveport smaller. But what if I want a large system? How do I avoid the "rectangular waveguide modal excitation" error?