Published Nov 8, 2026 · Last updated Nov 8, 2026 · 4 min · IndieRF
Microstrip EM vs Hammerstad-Jensen: Z0 0.3% apart
A 1.5 mm microstrip on 0.8 mm FR-4 reads Z0 50.42 Ω and εeff 3.295 in a shielded-box EM solve at 1 GHz. Hammerstad-Jensen gives 50.58 Ω and 3.331, and most of the εeff gap is the box.
A 1.5 mm strip on 0.8 mm FR-4 (εr 4.4), solved in Planar EM, a method-of-moments simulator that puts the layout in a shielded box, reads Z0 50.42 Ω and εeff 3.295 at 1 GHz. Hammerstad-Jensen, with the Kirschning-Jansen dispersion term, gives 50.58 Ω and 3.331 for the same strip in open air. That is −0.3% on Z0 and −1.1% on εeff. For a uniform line the closed form is close enough to build, and the EM run is the cross-check.
The box takes most of the εeff gap
A solver of this kind needs walls. The box for this line is 12.5 mm wide, with a lid 8.8 mm above the ground plane, so 8 mm of air over the board. The walls and lid add capacitance through air, which puts a larger share of the field outside the FR-4 and lowers εeff. A 2-D quasi-static solve of the same box cross-section, with the same dispersion term added, gives 50.26 Ω and 3.301. The box alone moves the closed form −0.6% on Z0 and −0.9% on εeff. Against that box reference, the EM solve is +0.3% on Z0 and −0.2% on εeff.
So most of the 1.1% is the enclosure. If your board goes into a shield can that close, the box number is the one to believe. Planar EM warns when the walls sit closer than 5h + W to the metal or the lid lower than 10h, because a tight box biases every number on the layout.
Scroll sideways for more columns
| Source | Z0 | εeff |
|---|---|---|
| Hammerstad-Jensen with Kirschning-Jansen dispersion, open space | 50.58 Ω | 3.331 |
| 2-D quasi-static solve of the same box, plus the dispersion term | 50.26 Ω | 3.301 |
| Planar EM, 50 µm cells, de-embedded | 50.42 Ω | 3.295 |
| Planar EM, limit of the mesh study | 50.25 Ω | 3.298 |
| Planar EM, 50 µm cells, raw wall ports | 48.02 Ω | 3.654 |
The mesh is part of the answer
The golden mesh uses 50 µm cells: 111 unknowns, well under a second in the browser. Halving the cell through 100, 50, 25 µm reads 50.63, 50.42, 50.32 Ω. Each step is about half the one before, which is first-order convergence: the charge crowds toward the strip edges and a uniform cell resolves that crowding slowly. Extrapolating the cells, and adding the smaller move from refining along the line, lands on 50.25 Ω, 0.01 Ω below the box reference. The 50 µm answer sits +0.3% above its own limit.
Scroll sideways for more columns
| Cell | Unknowns | Z0 | εeff |
|---|---|---|---|
| 100 µm | 85 | 50.63 Ω | 3.2934 |
| 50 µm | 111 | 50.42 Ω | 3.2951 |
| 25 µm | 137 | 50.32 Ω | 3.2960 |
| Limit | 50.25 Ω | 3.2985 |
Controlled impedance is ±10% at most fabs, ±5 Ω on a 50 Ω line. The 0.16 Ω between EM and the closed form is about a thirtieth of that window, and nobody will ever measure it on a coupon. Run a finer cell when you want the third digit, and keep the 50 µm answer when you want the layout.
What 1% of εeff costs in delay
Delay goes as √εeff. EM reads 153.79 ps/in and the closed form 154.63 ps/in, both close to the familiar 150 ps per inch for FR-4 microstrip. The 0.84 ps per inch adds up to 3.3 ps over 100 mm, which is 2.9° at 2.4 GHz. That matters for a phase-matched pair or a delay line trimmed to the degree. A 50 Ω feed to an antenna can ignore it.
Check the reference planes first
Read Z0 and εeff straight from the raw wall-port S-parameters and the same line gives 48.02 Ω and 3.654, −4.5% and +10.7% against the box reference. The port at the wall carries its own reactance. Planar EM removes it with two line standards solved on the same mesh, so the reference planes sit on the walls and the Touchstone file it downloads is already de-embedded. When a box solver and a closed form disagree by several percent, look at where the planes are before you blame the mesh.
Ends and tees are where EM pays off
On a straight line the closed form already knows the answer. At a discontinuity the closed forms thin out. An open-ended 1.5 mm line on the same board, resonating near 2 GHz, reads 2.00900 GHz in Planar EM, 2.00720 GHz from the Kirschning-Jansen-Koster open end in the same box (+0.09%), and 2.00892 GHz from openEMS, an independent FDTD solver run offline on the same box (+0.004%).
A shunt open stub is a harder test. Drawn length plus one open end puts the notch at 1.998 GHz. Planar EM finds 2.036 GHz, 38 MHz higher, because the tee junction changes the stub's electrical length. The drawn-length estimate is −1.9% from EM. Hammerstad's tee model brings the closed form to 2.032 GHz, −0.2% from EM. Without a tee model, that 1.9% is the error you etch into the board.
Every number here is lossless, with zero-thickness metal. Copper thickness lowers Z0 a little and solder mask raises εeff. The Transmission Line Calculator field solve includes both; Planar EM treats the metal as a sheet. Past 2,500 unknowns a browser run shows its time and memory before it starts. The full set of cases, references, and mesh studies is on the validation page, and any result opens in the S-parameter Viewer from its Touchstone file.
References
- E. Hammerstad and Ø. Jensen, "Accurate Models for Microstrip Computer-Aided Design," IEEE MTT-S International Microwave Symposium Digest, 1980, pp. 407–409.
- M. Kirschning and R. H. Jansen, "Accurate model for effective dielectric constant of microstrip with validity up to millimetre-wave frequencies," Electronics Letters, vol. 18, no. 6, 1982.
- M. Kirschning, R. H. Jansen and N. H. L. Koster, "Accurate model for open end effect of microstrip lines," Electronics Letters, vol. 17, no. 3, 1981.
- E. Hammerstad, "Computer-Aided Design of Microstrip Couplers with Accurate Discontinuity Models," IEEE MTT-S International Microwave Symposium Digest, 1981.
- J. C. Rautio and R. F. Harrington, "An Electromagnetic Time-Harmonic Analysis of Shielded Microstrip Circuits," IEEE Transactions on Microwave Theory and Techniques, vol. 35, no. 8, 1987.
- openEMS 0.37.0, run offline as reference data for the open end.
Questions? Contact
Discussion
No comments yet. Start the thread with a measurement, a correction, or a worked example.