EMMA-FP 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-FP-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, with multiple device instances per model carrying per-device fingerprint priors. The rfgen FingerprintConfig pins CFO (carrier frequency offset), IQ imbalance, and phase noise per unit (cfo_hz_range, iq_imbalance_db_range, phase_noise_dbc_hz_range).

  • Receiver array: 8-element ULA (uniform linear array), half-wavelength spacing (spacing_lambda: 0.5), via the rfgen MultiRXConfig.geometry member ULA. 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 realized device instance; LabelExtractor maps emitters[].device_id to the device-identity label.

  • SNR (signal-to-noise ratio) regime: drawn from the rfgen channel.snr_db_range.

  • Sample counts: approximately 60k scenes total.

  • Splits + seeds: train / dev / holdout seed ranges pinned at release; the holdout is a frozen secret seed, never published. OOD axis: LEAVE_ONE_UNIT_OUT (every capture of one device held out together, so device identity is isolated from capture session).

Real-capture counterpart

EMMA-REAL-OOD-v0.1. Real-capture device fingerprints are not wired at v0.1 (synthetic only).

Notes

  • The confound fix: existing RF fingerprinting sets confound device identity with capture session, because each device is recorded in one session on one receiver. EMMA’s leave-one-unit-out protocol and per-device FingerprintConfig randomization are designed to isolate device identity. The recipe must also randomize the receiver front-end across units before scoring, or the score measures the receiver chain rather than the device.

  • Construct validity is entirely conditional on a rigorous leave-one-unit-out recipe; see Datasets validation.

References

  • DARPA RFMLS program, 2017 to 2021. The capability-without-public-benchmark precedent for fingerprinting. (verify program record)

  • Bihl, Bauer, and Temple, “Feature representation for RF fingerprinting,” 2017. SEI literature. (verify)

  • “No Radio Left Behind,” 2019. Cross-session device identification; the leave-one-unit-out motivation. (verify)

See Also