grazing incident or incidence

I am currently working on a problem which requires simulating a plane wave excitation at low angles of incidence (theta>75 degrees). I have set up the typical HFSS periodic unit cell model with PML on top and below my airbox that contains my structure. I get fine results until about theta=65deg beyond which I must manually mesh at a very tight tolerance to get a good solution. Even then the solution is only accurate to about 75 deg. Beyond that I have not had any luck. I think this is probably a numerical dispersion problem since the wave has to 'travel' a long way to get to the structure. Any ideas on how to get around this without such a high density mesh?

Thanks.

very interested in your project, may i have your model?
I have working on the problem which antennas built on a large object.
contact with faintt1980@hotmail.com
Best regard!!!