F
Member
【Editorial Discussion】
Two established field solvers giving different resonant frequencies or Q for the same board is a regular design-review argument. Before declaring one tool “wrong,” it is worth asking whether the two models are actually the same problem: mesh at the metal edges, port calibration, and the material description often differ more than the solver kernels do.
1) Align the geometry and the metal treatment first. Differences in corner radius, trace etch factor, via padstacks, or whether a solder mask is included will move a resonance. Export a common intermediate geometry, or at least a checklist of what was simplified in each tool.
2) Ports and de-embedding must be defined the same way. Wave ports, lumped ports, and different reference-plane shifts produce different apparent resonances even when the cavity is identical. If a thru or a simple known resonator already disagrees, stop debugging the product and fix the port setup.
3) Material models are a frequent silent mismatch. A constant Dk/Df versus a frequency-dependent model, roughness models, and whether copper is treated as a thick metal with finite conductivity will shift both frequency and loss. Write the material assignment next to the plot.
4) Mesh convergence should be shown, not assumed. A resonance that still moves when the mesh is refined is not yet a solver-to-solver discrepancy. Refine both models until each is stable, then compare.
Open question: when your last two solvers disagreed, what actually closed the gap — mesh, ports, or materials? Have you had a case where both simulations were “converged” and a measurement later showed a third answer? Please describe how you decide which model to trust.
Two established field solvers giving different resonant frequencies or Q for the same board is a regular design-review argument. Before declaring one tool “wrong,” it is worth asking whether the two models are actually the same problem: mesh at the metal edges, port calibration, and the material description often differ more than the solver kernels do.
1) Align the geometry and the metal treatment first. Differences in corner radius, trace etch factor, via padstacks, or whether a solder mask is included will move a resonance. Export a common intermediate geometry, or at least a checklist of what was simplified in each tool.
2) Ports and de-embedding must be defined the same way. Wave ports, lumped ports, and different reference-plane shifts produce different apparent resonances even when the cavity is identical. If a thru or a simple known resonator already disagrees, stop debugging the product and fix the port setup.
3) Material models are a frequent silent mismatch. A constant Dk/Df versus a frequency-dependent model, roughness models, and whether copper is treated as a thick metal with finite conductivity will shift both frequency and loss. Write the material assignment next to the plot.
4) Mesh convergence should be shown, not assumed. A resonance that still moves when the mesh is refined is not yet a solver-to-solver discrepancy. Refine both models until each is stable, then compare.
Open question: when your last two solvers disagreed, what actually closed the gap — mesh, ports, or materials? Have you had a case where both simulations were “converged” and a measurement later showed a third answer? Please describe how you decide which model to trust.
