FXC-4546: add baseband source time classes for transient simulations

[dmarek-flex]

Note

Medium Risk
Introduces new SourceTime variants and updates schema/type unions used broadly in simulation validation, which could affect source serialization/validation and downstream integrations relying on SourceTimeType.

Overview
Adds new baseband (no-carrier) SourceTime options for transient RF workflows: BasebandStep, BasebandGaussianPulse, BasebandRectangularPulse, and BasebandCustomSourceTime (including from_values, end_time, and FFT-based frequency_range).

Plumbs these new source times through the public API (tidy3d.__init__, tidy3d.rf), schema discriminators (Simulation.json, ModeSimulation.json, TerminalComponentModeler.json), and type wiring by moving SourceTimeType into a new components/types/time.py module; also tightens validation by requiring positive dt for CustomSourceTime.from_values and tweaking the low-frequency error message for sources.

Extensive new tests validate baseband behavior (phase locked to 0, amplitude shapes, end_time/frequency_range, plotting, and full Simulation validation) plus CustomSourceTime dt edge cases.

^{Written by Cursor Bugbot for commit 79ef5bd. This will update automatically on new commits. Configure here.}

[github-actions]

Diff Coverage

Diff: origin/develop...HEAD, staged and unstaged changes

• tidy3d/init.py (100%)
• tidy3d/components/microwave/time.py (100%)
• tidy3d/components/source/base.py (100%)
• tidy3d/components/source/time.py (88.9%): Missing lines 110,115
• tidy3d/components/types/time.py (100%)
• tidy3d/plugins/smatrix/component_modelers/base.py (100%)
• tidy3d/rf.py (100%)

Summary

• Total: 149 lines
• Missing: 2 lines
• Coverage: 98%

tidy3d/components/source/time.py

Lines 106-119

  106         freqs = np.fft.rfftfreq(len(values), d=dt)
  107 
  108         peak = np.max(spectrum)
  109         if peak == 0:
! 110             return (0.0, 1.0)
  111 
  112         cutoff = np.exp(-(num_fwidth**2) / 2)
  113         above_cutoff = spectrum >= cutoff * peak
  114         if not np.any(above_cutoff):
! 115             return (0.0, freqs[-1])
  116 
  117         indices = np.where(above_cutoff)[0]
  118         fmin = float(freqs[indices[0]])
  119         fmax = float(freqs[indices[-1]])

---

[dmarek-flex]

Here are some nice pictures of the new baseband signals

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with Cloud Agents, enable Autofix in the Cursor dashboard.}

[dmarek-flex]

This one is basically ready to go. I opted for creating new classes to handles these signals with DC components, instead of modifying the existing ones to support freq0=0, I think the distinction is clearer this way. But open to your thoughts.

[George-Guryev-flxcmp]

Thanks, Damian — this looks solid! I have a few minor comments and suggestions.

[George-Guryev-flxcmp]

I believe the variable rise_time is misleading in this context: the rise time is typically a time it takes to go from 10% to 90%, while the variable rise_time is used here as sigma in step function below. The actual rise time is approximately 2.56 * sigma. I wonder one can rename rise_time to standard_deviation or something similar? Or alternatively use the actual rise time for this variable.

[George-Guryev-flxcmp]

I guess the default nominal value makes sense. However, I wonder if one could have a scenario where a simulation has multiple ports (say two) whose sources/waveforms should have a constant phase offset to each other? One can surely do phase scaling in post-processing, however, I wonder if you think this might be a useful behavior here (i.e. default to 0 but allow for any constant phase offset for a given waveform?)

[George-Guryev-flxcmp]

I believe this uses the RMS bandwidth; however, multiple alternative definitions exist. Making this explicit in the docstring would remove any ambiguity.

[George-Guryev-flxcmp]

All amp_time() methods give a type hint float for time. However, in most cases time is treated as a numpy array. I wonder you could use time: Union[float, ArrayLike] or something similar?