L
Admin
Full-wave sims are easy to start and hard to trust. Most bad HFSS conclusions I have seen were not wrong materials — they were under-converged meshes sold as final answers.
Before you refine forever
1. Fix the geometry hygiene first: tiny unintended gaps, overlapping metals, and sliver dielectrics create fake field concentrations and endless meshing.
2. Set boundaries and ports the way the fixture really behaves. A waveguide port on a microstrip launch without the right de-embed distance trains you to chase mesh when the port definition is the bug.
3. Seed the mesh on known critical features: gaps in filters, via fences, bondwire loops, connector pin gaps. Blind global refinement wastes tetrahedra on empty box air.
Convergence checks that actually mean something
- ΔS magnitude and phase between adaptive passes, not only a single |S11| number. Phase can still be moving when |S| looks flat.
- Field plots at the last two passes: if E hotspots jump location when the mesh changes, you are not converged.
- Power balance / residual error if the solver reports it. A pretty S-parameter plot with poor balance is a warning.
- Frequency sweep after adaptive meshing at the hardest frequency (often band edge or resonance), then confirm a few points with a finer mesh.
Cost control tactics
- Use symmetry planes when the layout truly has them.
- Start with a coarser target, lock a mesh that passes the checklist, then sweep.
- For tuning studies, converge once on the nominal, then use mesh reuse / same mesh for small parametric steps and only re-adapt when geometry topology changes.
- Compare against a second method on one golden case (circuit model, another solver, or measurement) before you trust a large DOE.
Stop rule I like: two consecutive adaptive passes with max |ΔS| below your design margin (for example 0.02–0.05 linear for matching work), stable hotspots, and a sanity check against a known limit case (shorted stub, empty waveguide, analytic patch estimate).
What ΔS threshold do you require before you call a filter or antenna sim "done"?
Before you refine forever
1. Fix the geometry hygiene first: tiny unintended gaps, overlapping metals, and sliver dielectrics create fake field concentrations and endless meshing.
2. Set boundaries and ports the way the fixture really behaves. A waveguide port on a microstrip launch without the right de-embed distance trains you to chase mesh when the port definition is the bug.
3. Seed the mesh on known critical features: gaps in filters, via fences, bondwire loops, connector pin gaps. Blind global refinement wastes tetrahedra on empty box air.
Convergence checks that actually mean something
- ΔS magnitude and phase between adaptive passes, not only a single |S11| number. Phase can still be moving when |S| looks flat.
- Field plots at the last two passes: if E hotspots jump location when the mesh changes, you are not converged.
- Power balance / residual error if the solver reports it. A pretty S-parameter plot with poor balance is a warning.
- Frequency sweep after adaptive meshing at the hardest frequency (often band edge or resonance), then confirm a few points with a finer mesh.
Cost control tactics
- Use symmetry planes when the layout truly has them.
- Start with a coarser target, lock a mesh that passes the checklist, then sweep.
- For tuning studies, converge once on the nominal, then use mesh reuse / same mesh for small parametric steps and only re-adapt when geometry topology changes.
- Compare against a second method on one golden case (circuit model, another solver, or measurement) before you trust a large DOE.
Stop rule I like: two consecutive adaptive passes with max |ΔS| below your design margin (for example 0.02–0.05 linear for matching work), stable hotspots, and a sanity check against a known limit case (shorted stub, empty waveguide, analytic patch estimate).
What ΔS threshold do you require before you call a filter or antenna sim "done"?