Jaguar2 thermal TX-power tracking (8822B/8821C)

[josephnef]

What

Jaguar2 was the only devourer chip family with no thermal TX-power tracking. HalJaguar2::apply_tx_power programs the efuse-calibrated TXAGC once at bring-up and it never moves again. On a sustained TX session (the OpenIPC long-range video case) the PA heats, gain drops, and on-air power drifts down with nothing correcting it — the droop that Jaguar1 (PowerTracking8812a) and Jaguar3 (Halrf8822c/e::pwr_track) both compensate.

This ports the vendor-faithful MIX_MODE power tracking for both variants.

How

A ~2 s tick (dedicated thread) per chip:

1. Reads the RF 0x42[15:10] thermal meter, 4-sample rolling average.
2. delta = |avg − efuse_baseline(0xBA)|, clamped to the table.
3. Looks up the per-variant / per-band delta-swing table (2.4 GHz + 5 GHz three sub-bands), transcribed from the vendor halhwimg88{22b,21c}_rf.c.
4. Writes the compensation as the vendor's get_mix_mode_tx_agc_bb_swing_offset_8822b split: coarse steps → TXAGC 0xc94/0xe94, remnant → BB scale 0xc1c/0xe1c via tx_scaling_table_jaguar.

• 8822B (2T2R): both PA chains driven from the single die thermal sensor (path B's 0x42 reads 0 — its sensor isn't enabled), each with its own swing curve; writes 0xc94[29:25].
• 8821C (1T1R): path A only; writes 0xc94[6:1].
• No LCK retrigger — the vendor #if 0s it for these chips.

The thread starts at the end of bring_up, so it covers both Init (RX) and InitWrite (TX-only, the sustained-TX case the RX-only DIG thread never reaches). A new _reg_mu serialises the tick's RF-window read/writes against SetMonitorChannel / FastRetune / the TX-power setters / GetThermalStatus.

Default on. DEVOURER_THERMAL_TRACK=0 disables it; an unprogrammed efuse baseline (0xFF) disables it automatically.

Validation

Build + ctest green (no warnings on the new code). Hardware-tested on a T3U (8822B) and CF-811AC (8821C):

• Functional soak (both variants, TX-only, max duty): thread starts, meter reads live and rises under heating, swing applied, TX alive, DEVOURER_THERMAL_TRACK=0 fully inert, clean shutdown.
• Two-adapter on-air (8822B → 8812AU, ch6/1M): 38 rx.txhit of the canonical beacon with tracking live.
• USRP power measurement at a matched hot state (the definitive proof): tracking ON delivers +3.74 dB more on-air power than OFF, matching the driver-logged swing index of 8 (≈ +4.0 dB expected, 0.5 dB/step) — the compensation reaches the antenna and restores exactly the PA droop that OFF leaves uncorrected.

New regression scripts: tests/jaguar2_thermal_track.sh (functional + off-inert), tests/jaguar2_thermal_track_onair.sh (two-adapter txhit), tests/jaguar2_thermal_track_sdr.py + tests/run_jaguar2_thermal_track_sdr.sh (SDR power proof).

Closes #184.

🤖 Generated with Claude Code