Follow-ups left open by the Jaguar3 wide-TX IGI floor (#484)

Offen
#485 0 Kommentare 0 Reaktionen 0 zugewiesene Personen Auf GitHub ansehen

Maintainer antworten meist innerhalb von 1 Tag

Dieses Issue hat noch niemand übernommen.

Bewertung

Schwierigkeit
5/5
Geschätzter Aufwand
Über eine Woche
Anfängerfreundlichkeit
15/100
Issue-Typ
Bug
Klarheit
Größtenteils klar
Aktivitätsstatus
Aktiv
Tech-Stack
c
Bereich
networking

Rechercherichtung

Start with the idle-vs-live register diff in docs/experiments/8822c-ht40-tx-stall.md and the wide_tx_igi_floor path that sets 0x35, using DEVOURER_WIDE_TX_IGI_FLOOR=0 as the control. Each bullet is a bench measurement or an open hardware question, so the work needs the Jaguar3 test units. Done means the RX-cost, 0x2e-0x32 bisect, 80 MHz and 2.4 GHz results are recorded in the experiment doc.

Vom Indexierungsmodell aus dem Issue-Text verfasst.

Beschreibung

enhancement

Open items from #484 (Jaguar3 40/80 MHz inject: wide_tx_igi_floor on both dies, 8822C DISDATAFB on wide frames), none of them blocking the merge:

  • The floor's RX-sensitivity cost is an estimate. The default 0x35 is 21 IGI steps over the DFS floor and 19-23 over the non-DFS window, quoted as "~1 dB per step" and never measured. This is the number an FPV uplink actually pays on the wide-transmitting side. Measure it: same transmitter at a fixed attenuation into a wide-TX session with the floor at 0x35 vs DEVOURER_WIDE_TX_IGI_FLOOR=0 (RX-only arming never happens, so the control is a session that sends one wide frame then listens), delivery vs attenuation, both dies.
  • No margin story behind the default. 0x35 is the lowest value that cleared the four units measured, not a margin over a characterized edge: one 8812CU clears at 0x24, the second hangs at 0x26 and airs ~1.5 k frames then wedges at 0x2c, and was never bisected between 0x2c and 0x35. Bisect that unit (0x2e/0x30/0x32, silent 144(-), MCS3/40, the 257 submitted / 8 failed / 0 aired shape as the fail criterion) so the default is "edge + N" rather than "the highest value anyone needed".
  • The real lever is unidentified. The hold is the receiver parking, on a quiet channel, in a gain state where the secondary 20 MHz never reads idle; any received traffic re-runs the AGC and clears it, DEVOURER_DIS_CCA (MAC CCA + EDCCA + sec20/40/80 gates) is inert against it, and the vendor phydm has no bandwidth-raised DIG bound (phydm_dig_abs_boundary_decision pins DFS at DIG_MIN_DFS at every width). The IGI floor is a workaround that buys TX at the cost of RX gain. The idle-vs-live register diff in docs/experiments/8822c-ht40-tx-stall.md (RF 0x08/0x58, CCK/AGC status 0x1a30-0x1a68) is where to look for a state the secondary CCA path reads; a lever that re-arms the AGC without raising IGI would make the floor unnecessary.
  • 80 MHz only partially validated. CU -> EU on 149(+) at VHT1SS_MCS3/80 with the old 0x26 floor ran 470 / 10 / 209; the 0x35 default has not been flown at 80 MHz on either bench. Re-run on a silent 80 MHz block with the default and with the floor off.
  • 2.4 GHz HT40 (ch 6). The CU drained on the TX side but the witness received none of it, before and after the change. Unexplained; not a 5 GHz-only feature claim until it is.
  • The 8812EU module's scattered send failures. On the second bench the BL-M8812EU2 drops single sends beside coex: tick failed (rtw_read: iostream error) in otherwise clean runs (13/8501 on 144(-), 64/7316 on busy 36(+)); the CU on the same hub never does. Not the stall shape, and probably that unit's USB wiring (bare module on a bench supply), but it has not been proven so: swap port / cable once and see whether the errors follow the module.

🤖 Generated with Claude Code

Vorherrschende Sprache
C
Sterne
56
Forks
29
Ø Merge
3 Std. 43 Min.
Gemergte PRs (30 T.)
75

Entwicklungsumgebung

Dieses Projekt bietet weder Dev-Container noch Dockerfile noch Beitragsleitfaden – die Einrichtung liegt bei Ihnen. Beginnen Sie mit der README; die allgemeinen Schritte stehen in unserem Leitfaden für den ersten Beitrag.

Erste Schritte

  1. Lesen Sie das ganze Issue und danach den Beitragsleitfaden des Projekts.
  2. Schreiben Sie ins Issue, dass Sie es übernehmen — das erspart doppelte Arbeit.
  3. Forken Sie das Repository und arbeiten Sie in einem Branch.
  4. Öffnen Sie einen Pull Request, der die Issue-Nummer nennt.

Mehr aus OpenIPC/devourer

Alle Issues in OpenIPC/devourer

Ähnliche Issues

Weitere Issues zu C

Neue Issues direkt in Ihr Postfach

Eine kurze Übersicht über anfängerfreundliche GitHub-Issues.