EMMA-CSI v1

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-CSI-v1 covers the task, target domain, input, labels, splits, and the download command; this recipe pins only the rfgen configuration. Source. rfgen (primary); CSI is a derived view computed from the rfgen channel, not the primary I/Q (in-phase and quadrature).

rfgen recipe

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

  • Emitters: the rfgen emitter_zoo group, comms family configured as 5G NR (New Radio) signals.

  • Receiver array: massive-MIMO (multiple-input multiple-output) array, 32 or 64 elements, via the rfgen MultiRXConfig group. Loaded into ArraySpec at load time.

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

  • Labeler + label extraction: the channel matrix is derived from the rfgen channel realization, not stored as the primary on-disk signal; LabelExtractor exposes the realized channel tensor as the target. The metric is NMSE (normalized mean square error) over the channel-tensor energy.

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

  • Sample counts: <to be pinned at v1>.

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

Real-capture counterpart

DeepMIMO is the external bridge data source for the CSI camp; a real-capture CSI counterpart is not part of the EMMA-REAL-OOD set.

Notes

  • CSI is a derived view, not the primary input. The raw multi-antenna I/Q remains the canonical record; the channel matrix is computed from the rfgen channel realization for this task only.

  • This is an on-ramp for the CSI camp (DeepMIMO / LWM (Large Wireless Model) lineage), not a head-to-head CSI-feedback competition. A dedicated CSI readout head is proposed to replace the interim angular regression head at v1.

References

  • Alkhateeb, “DeepMIMO: A Generic Deep Learning Dataset for Millimeter Wave and Massive MIMO Applications,” 2019, arXiv:1902.06435. The CSI-level data source for the CSI camp.

  • Djordjevic, Ali, and Alkhateeb, “LWM: A Foundation Model for Wireless Channel Data,” 2024, arXiv:2411.08872. The CSI-native foundation-model lineage. (verify)

  • 3GPP TR 38.843, “Study on AI/ML for NR.” Names CSI feedback as an AI/ML use case without dataset or baseline. (verify revision-year)

See Also