Changelog#

All notable changes to this project will be documented in this file. The format is based on Keep a Changelog and follows Semantic Versioning.

v0.4.0 - 25 Sep 2026#

A structures-and-export release. It adds a printed inverted-F antenna, building blocks for your own structures, an experimental wave port for the FEM backend, and dark and light plot themes. The Gerber and STEP exports now handle vias, their antipads, inner layers and transformed objects well enough to send a design to a board house. A round of FDTD mesher fixes makes simulations smaller and more accurate, and scikit-rf is no longer a dependency.

Added#

  • InvertedFAntenna / InvertedFAntennaParams: a printed inverted-F antenna designed from a target frequency, with an example, a tutorial and API docs (fd9f964, 961cdb5, 7f8e3ac)

  • GenericParams / GenericStructure: build your own structure from ready-made pieces – traces, bends, tapers, stubs, vias, coplanar waveguides and ports – with an example and a “building a custom structure” tutorial (c1e2be2, 9bd6fe7)

  • A SimParams subclass can supply substrate_width_mm and substrate_length_mm as plain fields instead of properties (32fcdd5)

  • Dark and light plot themes, use_dark_theme() and use_light_theme(), with an API docs page. Plots now default to dark (1035fa8, 5adff82)

  • FEM: an experimental wave port, FEM_port_type="waveport". It solves the mode of the line at the port – microstrip, coplanar waveguide and so on – instead of driving one fixed field direction. The port sits on the edge of the simulation and sizes itself from the trace and substrate; FEM_waveport_width_mm and FEM_waveport_height_mm set that size directly, and FEM_port_mode_* pick the mode and override its impedance or permittivity. Lumped ports stay the default (9f000ab, ffe29f1, 41b731f)

  • FEM: FEM_boundary="pec" closes the model in a shielded box, for transmission lines, filters and other structures that do not radiate. The backend also reports how much signal a line loses (41b731f)

  • FEM: FEM_mesh_freq sets the frequency the mesh is sized at; see Changed (81d6483)

  • FEM: FEM_air_pad_mm also takes [x, y, z] or three [low, high] pairs, so each face of the air box gets only the air it needs. Zero puts the boundary right on a ground plane. A single number works as before (76a9b63)

  • FEM: FEM_max_solve_points caps the extra solves the sweep spends correcting a physically impossible curve (see Fixed) (43f1335)

  • FEM: the progress output shows the number of frequencies to solve, the time each solve took, the elapsed time, the degrees of freedom and the peak memory (3423bba)

  • simulate_model() records an NF2FF box, so a standalone model can plot its radiation pattern and directivity (a20d7cc, 7839433, b669af6)

  • simulation_box on SimParams sets the FDTD simulation box as sizes or as bounds. The mesher and the field dump both use it; left as None, both derive it from the geometry as before (39cb572)

  • export_touchstone, simpleEMS’s own Touchstone writer and reader. Its data lines match the old scikit-rf output byte for byte. SimTools.export_touchstone() takes a new s_matrix argument to export every S-parameter of a multi-port network, not only S11 and S21, and returns the path of the file it wrote (6a48784)

  • Via support in the STEP export and, through it, in the FEM backend (d9cbaf0)

  • The simpleEMS logo in the README and the docs header, and figures redrawn on a dark background (4139ef5, 2784643, 3593620, b317cbd, e99e912)

Changed#

  • Breaking: removed SimTools.run_all_post_processing(). Call the plot and export methods you need directly (b7de19a)

  • Breaking: the Gerber export writes one file per copper layer, chosen by the z position of each metal, plus an Excellon drill file and a board outline, instead of writing only the top layer (6504ee6)

  • Breaking: substrate_cells defaults to 7 instead of 4, and values below 2 raise an error. The mesher places that many evenly spaced z-lines through the substrate, counting both faces (07a4531)

  • FEM: the mesh and the air box follow main_freq instead of the ends of freq_range. Widening the range to see more of a plot no longer makes the mesh finer and the air box bigger (81d6483)

  • The microstrip line’s board is now as long as the trace, with a port on each edge, instead of a dielectric wavelength of padding around it: a 3 mm trace used to sit on a 120 x 133 mm board (6d906be)

  • The band-pass filter and the IFA short their stubs to ground with real vias instead of boxes, so the shorts now reach the Gerber export (b76030a, 4e5053e)

  • The FDTD simulation box pads by a full wavelength instead of half of one, so the NF2FF box encloses the whole structure instead of cutting into the substrate (a2e93c5)

  • Dropped the scikit-rf dependency (59a9b0c)

Fixed#

  • FEM: a sweep could return a curve with more power leaving the structure than entered it, flagged only by a warning. The sweep now spends extra solves to correct itself, stops once they stop helping, and names the setting to change when it cannot. The curves it returns also track the real response more closely (43f1335)

  • FEM: boards with vias failed to simulate, because the mesher left a broken element on the via barrels (d6027e1)

  • FEM: on multilayer boards a port could stop just short of an inner ground plane, so the simulation saw a broken connection and reported a bad match (713ad93)

  • FEM: board loss came out about a fifth too high (41b731f)

  • FDTD: the auto mesher missed the narrow slot beside an inset patch’s feed, so the antenna came out badly mismatched. The mesher now places lines on the slot and spends fewer cells elsewhere, so the same simulation is smaller and faster (86895dd)

  • FDTD: above about 25 GHz the mesher packed five z-lines into the 35 um copper instead of one. The 60 GHz patch example now runs 2.3x lighter (3a3f933)

  • FDTD: polygons and curved bends produced so many fine mesh lines that the simulation could not run, and the IFA came out overmeshed (92191dd, 3169c24, fdbbe46)

  • FDTD: the mesher ignored a user-defined simulation box (39cb572)

  • FDTD: the thin-layer collapse could reduce a low-frequency substrate to a single z-line (07a4531)

  • simulate_model() on a model with no ports crashed while creating the NF2FF box instead of raising “No ports found” (42c8d79)

  • The Gerber and STEP exports placed rotated or translated objects in their original position, and the STEP export failed on a structure that reused a name (086fa36, 5c83cd6)

  • The antipad around a via never reached the Gerber layers, so a board made from them would have plated the via onto the plane it should pass through, and the STEP export left the planes whole. Both now cut the clearance, and the ground stitching around it stays connected (2843d16, b76df9f)

v0.3.0 - 09 Aug 2026#

Mostly a correctness release for the FEM backend: several results it reported were wrong, and the antenna radiation plots were the worst affected. If you have used the FEM backend for radiation patterns or gain, re-run those simulations. Sweeps are also faster, and STEP files can now be simulated with the FDTD backend too. This release also brings the project its first automated test suite and CI, so these results stay checked from now on.

Added#

  • STEP-file import for the FDTD backend (simulate_step_FDTD), matching what the FEM backend already offered, with an inset-fed patch example

  • add_fdtd_setup() adds simulation settings (frequency range, boundary conditions, timestep, end criteria) to a geometry-only CSXCAD file, so a model saved by CSX.Write2XML() can be simulated. It also changes the frequency band of an existing model without rebuilding its geometry

  • simulate_model(freqs=...) to report results over a frequency band you choose, rather than the one built into the model

  • add_field_dump(dump_freq=...) lets you pick the frequency a frequency-domain dump records; it defaults to the model’s main frequency

  • FEM: FEM_air_pad_mm to set the air padding around a structure directly, instead of letting it scale with wavelength. Useful for filters and other non-radiating structures, whose air box need not grow with a wide sweep

  • FEM: simulate_step_FEM() now shows the meshed geometry with PyVista after meshing, so you can check the mesh before the sweep runs. show_mesh turns it on or off, and mesh_style and theme match SimTools.write_and_show_structure()

  • Documentation: solver-backend and STEP-export sections in the README, and a geometry-only model section in the standalone model tutorial

  • A first automated test suite and CI workflow. Tests cover the design formulas, both solver backends, the exporters, the CLI and the plots, and the examples now fail the build if the API moves under them. Tests that need openEMS or GetDP installed skip themselves, so the rest still run anywhere. CI runs the fast tests on every push and the full suite on master

Changed#

  • Breaking: SimTools.export_stl() now takes sim as its first argument. STL files are built directly with CadQuery rather than by launching AppCSXCAD, so the export no longer needs that program installed

  • Breaking: FEM and FDTD settings now live on SimParams as FEM_* and FDTD_* fields, instead of being passed to setup_simulation() (num_FEM_solve_points is now FEM_num_solve_points)

  • Breaking: removed the port_type / FEM_port_type "wave" option. It was never a real wave port, and a genuine one, prototyped and measured, performed worse than the lumped port it would have replaced. Ports are lumped-only for now

  • FEM: every port at a frequency is now solved together rather than one at a time. This roughly halves a two-port sweep, and saves more the more ports there are. Results are unchanged

  • FEM: each material is now meshed to the detail its own wavelength needs, so the substrate is resolved properly without spending elements on empty air

  • FEM: a sweep whose results are physically impossible is now reported as such, rather than returned as a curve with a false resonance in it

  • FEM: the formulation now uses the e^{+jωt} engineering time convention that openEMS uses, instead of the physics e^{-jωt} one. A phase, and an inductance or capacitance read off the Smith chart, now mean the same thing whichever backend produced it

  • simulate_model() and add_fdtd_setup() now write to Sim_Path when given no output path, as the other modules do. simulate_model()’s output_path is optional as a result

  • Modules renamed to make clear which backend they belong to: mesh.py to fdtd_mesh.py, standalone_model.py to fdtd_standalone_model.py, and export_step.py to export_cad.py

  • The FEM modules’ documentation was rewritten to a consistent style

Fixed#

  • FEM antenna gain and radiation patterns were wrong in four separate ways, each of which alone could shift a pattern or its level:

    • the far field was computed in the wrong time convention, which affected every radiation plot and the directivity read off it

    • radiated power was measured in a way that reads near zero behind a PML boundary, leaving those gain plots around 14 dB low

    • with a symmetry plane, only half the antenna was accounted for, and part of what was measured sat outside the simulated region

    • efficiency was taken from the strongest point of the sweep rather than the frequency the pattern was computed at

    A half model of the 24 GHz patch example now matches the full model to 0.25 dB, and a PML run matches a Silver-Muller one to 0.16 dB. Loss in lossy conductors is counted for the first time as part of this

  • FEM: when the usual far-field pattern file cannot be read and the fallback reads it point by point, each sample’s coordinates were being mixed up, so every sample pointed the wrong way. The fallback now reads them correctly

  • FEM S-parameters between ports that differ from each other – in substrate thickness or reference impedance – were wrong, which could also make a reciprocal structure look non-reciprocal. Ports that match each other, as in every shipped example, were unaffected

  • FEM S21 was reported with the wrong sign of phase, giving negative group delay

  • FEM: group delay had the wrong scale as well, dielectric loss was measured over the wrong region, and a solid whose name sounded like a port (say port_feed_1) could be classified as metal, silently dropping the excitation. All three are corrected

  • FEM: a sweep over a single frequency used to crash, and when it did not it solved that frequency five times and threw four of the results away, because the seeding step spread its points across a band with no width. A single-point sweep now solves once

  • FEM simulations no longer reuse a stale mesh or solver setup after the geometry or the settings change, so sweep and optimise loops are correct

  • Box-shaped STEP solids are rebuilt as real boxes rather than approximated by flat faces, restoring the mesh detail at metal edges and with it the resonance the model predicts

  • The default output path when none is given, for both backends, and a relative output_path is now made absolute before running, so openEMS no longer fails with a confusing error about its working directory

  • simulate_model() on a model with no simulation settings now says what it found and points at add_fdtd_setup(), and an unsupported excitation now gives a clear error instead of a TypeError. It also accepts a pathlib.Path for the model file, as documented, not only a plain string

  • SimTools.run_all_post_processing() stopped partway through: it passed the wrong value to the STL export and errored out before ever writing the Touchstone and Gerber files. It now finishes everything it promises

  • Errors no longer dump local variables into the traceback, where they were mostly noise

  • simpleems install openems no longer reports a failed install after a successful build. The Python bindings were being installed into a virtual environment that nothing activates, and the health check ran before the freshly installed binaries were on PATH. AppCSXCAD, the optional Qt viewer, is now reported but no longer fails an install on a machine without Qt. On Windows, --force now really replaces the old files instead of leaving them behind

  • Documentation CI build, the demo screenshots that PyPI showed as broken, the project’s GitHub link in the docs header, and the description of what run=False does when importing a STEP file (it reads back results and fails when there are none, rather than stopping before the solve)

0.2.0 - 25 Jul 2026#

Added#

  • New GetDP-based FEM (finite-element) solver backend as an alternative to the FDTD (openEMS) backend, with shared FDTD/FEM modules unified for dual-backend simulation

  • simpleems install getdp CLI command to install the GetDP solver binary

  • Documentation: GetDP installation guide, fem_backend API reference, and a “Standalone FEM STEP Model” tutorial

Fixed#

  • Updated examples for the add_field_dump/compute_sim_data signature changes

  • Fixed README installation, PyPI, and demo links

  • Fixed Sphinx docs CI build (mocked heavy/native dependencies for a docs-only build) and added GitHub Pages deployment

0.1.0 - 11 Jul 2026#

  • Initial public release