GetRxQuality() — runtime RX link-quality feed + passive noise floor

[josephnef]

Why

Seeded by Fluke (BlackFox's experimental alink replacement): it "measures its environment and builds its own optimal txprofile" from richer driver telemetry than alink's RSSI/SNR — SNR/EVM, a noise-floor number, and PA/LNA-oversaturation. Its sharpest insight is self-jamming (Paddy's 10-ft-indoors): benching at high TX power raises the receiver's noise floor → SNR collapses → the link jams itself; the manual fix is "decrease txpower until NF returns to normal", which needs the driver to report the floor.

devourer had the pieces but not the feed: per-frame RSSI/SNR/EVM only on the Packet callback, the windowing + LinkHealth verdict only in demo/main.cpp, and no noise-floor metric at all. A controller linked against the library (fluke_gs) had to scrape stdout.

What

• src/RxQuality.h — RxQuality snapshot + a thread-safe RxQualityAccumulator (the demo's RxAgg promoted, plus a passive noise-floor term) + build_rx_quality() fusing the frame aggregate, GetRxEnergy, and classify_link_health once.
  • Passive noise floor: nf_dbm = rssi_dbm − snr_db per OFDM/HT frame — the effective floor the receiver sees. TX self-jam drops SNR while RSSI holds, so it rises. Works on all three families incl. Jaguar3 (no background DIG, so IGI can't track the floor).
• IRtlDevice::GetRxQuality() — the runtime feed; subsumes GetRxEnergy (calls it internally, consumes the FA/CCA delta). Each device owns an accumulator, feeds it per decoded frame in the RX loop, returns build_rx_quality(...).
• demo: DEVOURER_RXQUALITY=1 emits <devourer-rxquality> (dogfoods the API; existing <devourer-energy>/<devourer-linkhealth> untouched).

Validation

• Unit (RxQualitySelftest, ctest + mingw list): aggregate math, the NF formula against a synthesized saturation tuple, EVM/NF present-gates, delta-reset.
• On-air (tests/rx_noise_floor_onair.sh): the feed is valid across a TX-power sweep and the noise floor is self-consistent on real frames — worst |nf−(rssi−snr)| = 1.0 dB — with SATURATED firing at high power.

txidx | frames | nf_dbm | rssi_dbm | snr_db | verdict
   8  |  9362  | -76.2  |  -55.0   |  20.2  | HEALTHY
  20  |  7576  | -70.2  |  -51.0   |  19.5  | HEALTHY
  34  |  7432  | -64.8  |  -46.0   |  19.0  | SATURATED
  48  |  2258  | -91.1  |  -69.5   |  21.5  | SATURATED
  63  |  2575  | -91.1  |  -70.0   |  21.5  | SATURATED

Bench caveat (measured, honest): the bench can't produce a large controllable NF excursion — the wanted beacon is ~−50 dBm inches away, and neither TX self-jam nor a co-located B210 AWGN at 78 dB moves the 8822EU's SNR (its AGC absorbs it) — the same near-field/front-end limit the SDR-power + dis_cca work hit. The unit test covers the NF math directly; a large controllable swing needs a real far-field link.

Full all-chips build + J1-only subset + ctest (10/10) green. M2 (the frame-free active absolute noise floor, measure-first) builds on this struct next.

🤖 Generated with Claude Code