EMMA-REAL-OOD 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-REAL-OOD-v0.1 covers the purpose, capture sources, splits, and how the subset is used; this recipe pins only the adapter configuration.

Capture sources

This is not an rfgen recipe. Real captures are ingested through the RealCaptureAdapter ABC, with three concrete implementations:

  • ColosseumAdapter for the Colosseum testbed.

  • PowderAdapter for the POWDER testbed.

  • CosmosAdapter for the COSMOS testbed.

  • Arrays: 4 to 8 element arrays, matching what each testbed records.

  • Carrier / BW: various; each testbed contributes its own bands and bandwidths.

  • Sample counts: approximately 8k captures.

  • Split: dev only. Real captures are for fidelity measurement, never for training or official holdout scoring.

Synthetic counterpart

Paired with every synthetic recipe: EMMA-LOC-v0.1, EMMA-DRONE-v0.1, EMMA-FP-v0.1, and EMMA-AMC-v0.1. The sim-to-real gap is computed between a synthetic-trained model’s score and its score on these captures.

Notes

  • Because synthetic-only benchmarks inherit their generator’s blind spots, this subset is the load-bearing check on whether synthetic gains transfer. The sim-to-real gap is reported alongside every synthetic score, never folded into the aggregate.

  • The RF-Analyzer result that sim-to-real generalization breaks in low-SNR (low signal-to-noise ratio) and OOD (out-of-distribution) regimes is the load-bearing caveat: the expected failure region is low SNR and out-of-distribution.

  • Adapter coverage is uneven across tasks (drone-RF and device-fingerprint real captures are scarce); coverage gaps are flagged per task in Datasets validation.

References

  • Colosseum (Northeastern) RF testbed. (verify citation)

  • POWDER (Powder Platform for Open Wireless Data-driven Experimental Research) testbed. (verify citation)

  • COSMOS (Cloud-enhanced Open Software defined Mobile wireless testbed for city-scale deployment) testbed. (verify citation)

  • “RF-Analyzer,” 2026, arXiv:2605.04676. Concedes sim-to-real generalization breaks in low-SNR and OOD regimes; the load-bearing sim-to-real caveat. (verify)

See Also