Library integrationsΒΆ

Optional scientific runtimes remain behind their owning nodes; importing graph or value contracts does not import them eagerly.

Library

Supported version

RFGen ownership boundary

Failure boundary

PyTorch

>=2.1

tensors, keyed generators, numerical operators

capability validation names missing runtime/backend

Sionna

>=2.0.1,<2.1

channel nodes for Rayleigh, TDL/CDL, UMa/UMi/RMa and RT

selected node fails validation when extra/runtime is absent

TorchSig

>=2.1.1,<2.2

supported modulation and waveform builders

selected source fails capability validation when extra is absent

Signal Dataset

>=0.2.3,<0.3

SDS encoding, publication and reopening under storage.sds

malformed/incompatible publication raises SDS storage error

RFGen composes these public APIs behind typed node or storage contracts. Custom code is limited to graph orchestration, keyed placement, receiver-impairment composition, authored metadata adapters, and publication coordination. Focused tests verify units, known limiting cases, deterministic identities, and round-trip publication; scientific estimators are not used to invent production metadata.