R
Member
A single converged-looking S-parameter plot is not evidence that the model is right. Two verifications matter most.
1) Mesh convergence: run the same model with at least three mesh/adaptive-pass settings and plot the quantity of interest (S21 at band edges, resonance frequency, or impedance) against pass count. If the resonance moves by a few percent between the last two passes, the result is not converged. Record the final pass count and delta in the report. Local mesh refinement around gaps, vias and thin traces is usually more efficient than refining the whole volume.
2) Port setup and de-embedding: a wave port that is too small, or a lumped port with an ill-defined ground reference, adds a fictitious reactance. Always state the port type, dimensions, calibration plane and whether de-embedding was applied. Compare a de-embedded result against a simple analytic case (e.g. a uniform microstrip line of known impedance) before trusting the complex structure.
Other frequent issues:
Material definitions at the wrong frequency or with the wrong loss model
Radiation boundaries placed too close to the structure
Ignoring solder mask and surface roughness when working at higher frequencies
Comparing simulated results to VNA measurements without moving both to the same reference plane
Suggested deliverable: a short table with mesh setting, pass count, resonance frequency, peak |S11|, solve time, plus a plot of the convergence curve and a photo/sketch of the fixture used for measurement.
Takeaway: simulation credibility comes from documented convergence and a matched reference plane, not from a smooth plot.
1) Mesh convergence: run the same model with at least three mesh/adaptive-pass settings and plot the quantity of interest (S21 at band edges, resonance frequency, or impedance) against pass count. If the resonance moves by a few percent between the last two passes, the result is not converged. Record the final pass count and delta in the report. Local mesh refinement around gaps, vias and thin traces is usually more efficient than refining the whole volume.
2) Port setup and de-embedding: a wave port that is too small, or a lumped port with an ill-defined ground reference, adds a fictitious reactance. Always state the port type, dimensions, calibration plane and whether de-embedding was applied. Compare a de-embedded result against a simple analytic case (e.g. a uniform microstrip line of known impedance) before trusting the complex structure.
Other frequent issues:
Material definitions at the wrong frequency or with the wrong loss model
Radiation boundaries placed too close to the structure
Ignoring solder mask and surface roughness when working at higher frequencies
Comparing simulated results to VNA measurements without moving both to the same reference plane
Suggested deliverable: a short table with mesh setting, pass count, resonance frequency, peak |S11|, solve time, plus a plot of the convergence curve and a photo/sketch of the fixture used for measurement.
Takeaway: simulation credibility comes from documented convergence and a matched reference plane, not from a smooth plot.
