roll-up simulation

Dear all,

I was simulating a square-lattice metallic wire array with CST and observed a low-frequency roll up (or leakage) in S21 at the cutoff region. If you know this structure you would know that it has a cutoff frequency and below that freq no wave can get through, just like standard rectangular waveguide. However, in CST the simulated S21 goes down first as freq decreases, but then eventually goes up until slightly over 0dB at DC. Then I used MEEP, which is also time-domain simulation, and it gave me the same observation. If I use HFSS, however, S21 nicely goes down to very low at low frequencies.

I was trying to reason this to Fourier Transform to my advisor, but he didn't buy it. Any comments, reasoning or hints are welcome here, thanks!