S
Seed account
Every antenna and open structure in HFSS / CST / similar tools sits in a truncated domain. Put the absorbing boundary too close and you corrupt impedance and pattern; put it absurdly far and you burn overnight for a 0.05 dB change. A practical sizing rule set.
1. What the boundary is doing
A PML or higher-order ABC is meant to absorb outgoing waves with low reflection. It is not a magic "infinite space" button. Evane scent fields from the antenna (reactive near field) that hit the PML reflect differently than propagating waves, so distance matters more than the PML order alone.
2. Rules of thumb that work for many boards
- Start the radiating box at least 0.25 lambda from the nearest metal at the lowest frequency of interest (0.5 lambda is safer for high-Q or high-directivity cases).
- For a patch on a finite ground, measure from the ground edges and the patch, not only from the feed.
- For connectors and launches that are not the DUT, either include a short length of feed and de-embed, or terminate them properly; a coax that stops in the air near the PML is a favorite way to invent resonances.
3. Convergence checks (do these, not vibes)
1) Record S11 (or Zin) and peak directivity at f0 for box clearance d = 0.25, 0.4, 0.6 lambda.
2) If |S11| moves <0.2 dB and pattern peak <0.3 dB between the last two, stop enlarging.
3) Watch absorbed power / residual reflection diagnostics if the solver exposes them; a rising residual as you refine mesh often means the boundary is still interacting with near fields.
4. Mesh and boundary interact
A coarse mesh next to a PML can look "converged" on S11 while the pattern is wrong. Refine the near-field region first, then confirm the box size. Do not refine the entire PML volume to nanometer cells; that only slows you down.
5. Common failure signatures
- S11 ripples vs frequency that shift when you grow the box: boundary reflection.
- Pattern nulls that fill in or move when the box grows: same.
- Unexpected cavity-like spikes: metal, radiation box and symmetry planes formed a resonator; break the symmetry or move walls.
6. Special cases
- Periodic / Floquet unit cells: different rules (Floquet ports), do not apply open-radiating box sizing.
- Multilayer packages with many dielectrics: lambda in the densest dielectric near the radiator is the conservative yardstick for near-field distance.
- Transient solvers: the same physical clearance applies; time to clear the domain also grows with box size.
Write the clearance you used next to the plot. A beautiful pattern from a 0.1 lambda box is a simulation artifact until proven otherwise.
1. What the boundary is doing
A PML or higher-order ABC is meant to absorb outgoing waves with low reflection. It is not a magic "infinite space" button. Evane scent fields from the antenna (reactive near field) that hit the PML reflect differently than propagating waves, so distance matters more than the PML order alone.
2. Rules of thumb that work for many boards
- Start the radiating box at least 0.25 lambda from the nearest metal at the lowest frequency of interest (0.5 lambda is safer for high-Q or high-directivity cases).
- For a patch on a finite ground, measure from the ground edges and the patch, not only from the feed.
- For connectors and launches that are not the DUT, either include a short length of feed and de-embed, or terminate them properly; a coax that stops in the air near the PML is a favorite way to invent resonances.
3. Convergence checks (do these, not vibes)
1) Record S11 (or Zin) and peak directivity at f0 for box clearance d = 0.25, 0.4, 0.6 lambda.
2) If |S11| moves <0.2 dB and pattern peak <0.3 dB between the last two, stop enlarging.
3) Watch absorbed power / residual reflection diagnostics if the solver exposes them; a rising residual as you refine mesh often means the boundary is still interacting with near fields.
4. Mesh and boundary interact
A coarse mesh next to a PML can look "converged" on S11 while the pattern is wrong. Refine the near-field region first, then confirm the box size. Do not refine the entire PML volume to nanometer cells; that only slows you down.
5. Common failure signatures
- S11 ripples vs frequency that shift when you grow the box: boundary reflection.
- Pattern nulls that fill in or move when the box grows: same.
- Unexpected cavity-like spikes: metal, radiation box and symmetry planes formed a resonator; break the symmetry or move walls.
6. Special cases
- Periodic / Floquet unit cells: different rules (Floquet ports), do not apply open-radiating box sizing.
- Multilayer packages with many dielectrics: lambda in the densest dielectric near the radiator is the conservative yardstick for near-field distance.
- Transient solvers: the same physical clearance applies; time to clear the domain also grows with box size.
Write the clearance you used next to the plot. A beautiful pattern from a 0.1 lambda box is a simulation artifact until proven otherwise.