0 Fork 0

mirror of https://github.com/ruvnet/RuView.git synced 2026-10-23 15:13:03 -05:00

Files

T

ruv ab44fd798c feat(bfld): route HE through the capture pipeline + cluster runbook (ADR-365)

The cluster's real BE550/iPhone captures are HE (Wi-Fi 6), but decode_pcap and
the bfi_decode tool only routed VHT, so an HE capture would show 0 decoded.

- capture: new Report enum (Vht | He) with shared accessors; decode_pcap
  dispatches on action category (21 = VHT, 30 = HE).
- cbr: single-source dequant_phi/dequant_psi (pinned to WiPiCap by test),
  used by VhtBeamform and Report.
- features: ReportFeatures / motion_energy / motion_series work on either
  format; cross-format pairs are skipped (reconfiguration, not motion).
- bfi_decode: prints format, dims, subcarriers, bit widths, VHT/HE counts.
- docs/bfi-capture-integration.md: the capture->decode contract and command
  for any cluster node (Nexmon Pi, AX210, co-bound monitor).

Also fixes pedantic clippy hits in cbr/steering that an earlier, sloppy local
grep had hidden (doc backticks, long first doc paragraphs, mul_add, checked
casts). Verified: 28 lib tests; RUSTFLAGS=-D warnings test build; no-std
build; precise clippy -D warnings clean on all four touched modules; tool on
a mixed VHT+HE pcap decodes 4/4 (VHT=1, HE=3) at 20 reports/s.

Co-Authored-By: claude-flow <ruv@ruv.net>

2026-09-29 22:17:23 -04:00

2.7 KiB

Raw Permalink Blame History

BFI capture → decode integration (for the cluster)

The decoder side (ADR-365, PR #2053) is done and reference-validated. Any node that can capture beamforming feedback hands the decoder a radiotap pcap; nothing else is coupled. This is the seam.

Capture contract (what a capturing node must produce)

A pcap (classic .pcap, microsecond or nanosecond) with link type LINKTYPE_IEEE802_11_RADIOTAP (127) — i.e. 802.11 frames with a radiotap header. tcpdump -y IEEE802_11_RADIO -w out.pcap on a monitor interface.
It must contain management Action / Action No Ack frames carrying:
- VHT compressed beamforming: category 21, action 0, or
- HE compressed beamforming: category 30, action 0.
Beamforming feedback exists only on 5/6 GHz (VHT/HE). A 2.4 GHz capture has none. The client being sensed must be associated on 5/6 GHz and actively sounded (active downlink triggers sounding).

Capture-host options (any one is enough):

Nexmon on a BCM Pi (cognitum-v0 / Pi cluster) — ADR-123's intended host; the CBFR filter is the same code path as its CSI capture.
Intel AX210 in monitor mode (the card KIT used).
A co-bound monitor vif while associated (works where dedicated monitor RX is weak, e.g. the MediaTek MT7927 on ruvultra).

Decode + validate (what any node runs)

# branch feat/bfld-cbr-parser (PR #2053)
cargo run -p wifi-densepose-bfld --example bfi_decode -- capture.pcap

Prints a report-rate dashboard (decoded/sec, sequence gaps, dupes, failures), the first report's dimensions, and a motion-energy series. Library entry points for programmatic use:

wifi_densepose_bfld::capture::decode_pcap(&bytes) → VHT and HE reports (Report::Vht / Report::He) + ReportRateSummary
wifi_densepose_bfld::cbr::{parse_vht_action, parse_he_action} → structured angles
wifi_densepose_bfld::features::motion_series(&reports) → motion signal
wifi_densepose_bfld::steering::{reconstruct_v, unitarity_error} → V + the correctness gate

The validation gate (CLAIMED → MEASURED)

Reconstruct V from a captured frame and check unitarity_error(&v). A real capture whose V is unitary (< ~1e-6) confirms the decode end to end and is what promotes the pipeline from CLAIMED to MEASURED — no code change needed, just a real pcap. Decode math is already cross-checked against Wi-BFI and WiPiCap (bit widths, angle ordering, dequantization) and unitary on synthetic angles.

Not yet

HE MU/CQI and EHT (802.11be) return Unsupported — need a reference/capture.
Motion pairs are only formed between reports of matching format and dimensions (a VHT→HE switch is a reconfiguration, not motion).

2.7 KiB Raw Permalink Blame History