R
Member
We've all been there: the EM solver gives you beautiful S-parameters, you tape out, and the board comes back shifted by 200 MHz. Over the years I've settled on a mesh convergence routine before I trust any simulated result:
1. Run at least three mesh densities. I start with the solver's default adaptive mesh, then force one refinement pass, then another. If S11/S21 stop moving (I use <0.2 dB change across the band as my bar), the mesh is converged.
2. Watch the quantity you actually care about. Converging on S-parameters doesn't guarantee your radiated efficiency or coupling numbers converged too. Check convergence on the metric that drives your design decision.
3. Check the mesh, don't just trust the numbers. Look at where the solver put the dense cells. If the refinement is all in the wrong place (far from your coupled lines or the antenna feed), your "converged" result is converged on the wrong physics.
4. De-embed consistently. When comparing sim vs. measurement, make sure the reference planes match. Half the "sim doesn't match bench" stories I've debugged were actually de-embedding mismatches, not mesh problems.
5. Sanity-check against a closed form. A quick microstrip impedance calc or a coupled-line estimate takes two minutes and catches the gross errors — wrong substrate stackup, wrong units, port normalization mistakes.
My rule of thumb: if I can't explain why the mesh converged where it did, I don't tape out on it. What's your convergence criterion — and have you ever been burned by a "converged" simulation?
1. Run at least three mesh densities. I start with the solver's default adaptive mesh, then force one refinement pass, then another. If S11/S21 stop moving (I use <0.2 dB change across the band as my bar), the mesh is converged.
2. Watch the quantity you actually care about. Converging on S-parameters doesn't guarantee your radiated efficiency or coupling numbers converged too. Check convergence on the metric that drives your design decision.
3. Check the mesh, don't just trust the numbers. Look at where the solver put the dense cells. If the refinement is all in the wrong place (far from your coupled lines or the antenna feed), your "converged" result is converged on the wrong physics.
4. De-embed consistently. When comparing sim vs. measurement, make sure the reference planes match. Half the "sim doesn't match bench" stories I've debugged were actually de-embedding mismatches, not mesh problems.
5. Sanity-check against a closed form. A quick microstrip impedance calc or a coupled-line estimate takes two minutes and catches the gross errors — wrong substrate stackup, wrong units, port normalization mistakes.
My rule of thumb: if I can't explain why the mesh converged where it did, I don't tape out on it. What's your convergence criterion — and have you ever been burned by a "converged" simulation?
