EMMA-AMC v0.1

Warning

Pre-implementation. This page describes proposed contracts. Class signatures, parameter types, schema fields, and behavior are subject to change before code lands. Once implementation exists, content here will be regenerated from docstrings or sourced from running tests.

Dataset. EMMA-AMC-v0.1 covers the task, target domain, input, labels, splits, and the download command; this recipe pins only the rfgen configuration. Source. rfgen (primary).

rfgen recipe

  • rfgen commit: <rfgen-commit> (frozen at release; placeholder until pinned).

  • Emitters: the rfgen emitter_zoo group, comms family, using the full TorchSig-compatible comms modulation zoo (bpsk, qpsk, 16qam, 64qam, ofdm, and the rest of the standard set). rfgen composes TorchSig for these modulations; EMMA does not redefine them.

  • Receiver array: 8-element array, single-antenna compatible (the task is single-input single-output, SISO, solvable), via the rfgen MultiRXConfig group. Loaded into ArraySpec at load time.

  • Channel environment(s): E_URBAN and E_RURAL, pinned through the rfgen channel group.

  • Labeler + label extraction: the rfgen label group records the modulation class; LabelExtractor maps emitters[].modulation to the classification target.

  • SNR (signal-to-noise ratio) regime: wide SNR sweep drawn from the rfgen channel.snr_db_range; the OOD (out-of-distribution) fold holds out an unseen SNR regime under the SNR_REGIME axis.

  • Sample counts: approximately 200k scenes total.

  • Splits + seeds: train / dev / holdout seed ranges pinned at release; the holdout is a frozen secret seed, never published. OOD axis: SNR_REGIME (unseen SNR regime).

Real-capture counterpart

EMMA-REAL-OOD-v0.1 (fidelity reference, not a scored AMC target).

Notes

  • AMC is saturated, single-antenna-solvable, and its publisher states the data has known errata and is not used in its products. It is reported but excluded from the aggregated OOD-avg score. Treating AMC gains as benchmark progress re-imports a saturated task EMMA set out to move beyond.

  • Matched hyperparameter search collapses apparent model gaps on RadioML; see Hanna and Hussain.

References

  • O’Shea, Roy, and Clancy, “Over-the-Air Deep Learning Based Radio Signal Classification,” IEEE J-STSP 2018. The RadioML 2018 lineage. (verify)

  • DeepSig, “RadioML datasets,” datasets page (deepsig.ai/datasets). The publisher’s public note that RML2016 has known errata and is not used in DeepSig products. (verify URL)

  • Hanna and Hussain, “Robust Low-SNR Modulation Classification,” 2026, arXiv:2605.27673. Apparent model gaps collapse under matched hyperparameter search; grounds the saturation. (verify)

  • TorchSig. The signal-generation library rfgen composes for benchmark-compatible modulations.

See Also