AdvancedParaEQ: correct V2 band-stride and header length

The V2 parser was using stride [filter, gain_hi, gain_lo, freq_hi,
freq_lo] with a 3-byte header. Pcap traces of LGHUB ↔ G522 LIGHTSPEED
show the actual wire is [freq_hi, freq_lo, filter, gain_hi, gain_lo]
with a 1-byte header (just dir_echo; the centurion bridge omits the
slot_echo the protocol spec describes for getCustomEQ).

With the old stride, slot 0's 10-band response on G522 parsed as 9
bands: the highest band (19 kHz, +2.75 dB on the active preset) was
silently swallowed because byte alignment landed on what looked like a
freq=0 end-sentinel. The displayed bands were also shifted — 20 Hz was
missing from the front because the parser started one full band late.

See ~/ghub/solaar_0x020D_slot_investigation.md for the RE notes.
This commit is contained in:
Ken Sanislo 2026-05-10 17:21:09 -07:00
parent 298380b8d2
commit 7c73c88867
1 changed files with 22 additions and 22 deletions

View File

@ -23,16 +23,20 @@ V0/V1 wire format: 3-byte band stride [freq_hi, freq_lo, gain_i8],
gain is whole dB; getEQInfos returns 5 bytes [bandCount, dbRange,
caps, dbMin, dbMax].
V2 wire format: 3-byte header [direction_echo, slot_echo,
band_count_max], then N × 5-byte band stride [filter_type, gain_hi,
gain_lo, freq_hi, freq_lo], then 0..2 trailer bytes (opaque, ignored).
band_count_max is the device's max-bands capacity, NOT how many bands
are populated the parser consumes 5-byte chunks until <5 bytes
remain. Frequency u16 BE in Hz, with freq=0 marking end-of-bands.
Gain is **offset-binary**: raw 0..(steps-1) maps linearly to
gain_min..gain_max (so on G522 with steps=241 / gain=[-6..6], raw=120
= 0 dB). getEQInfos returns 13 bytes with gain bounds + step count,
format enum, XY-support flag, and onboard preset counts.
V2 wire format: 1-byte header [direction_echo], then N × 5-byte band
stride [freq_hi, freq_lo, filter_type, gain_hi, gain_lo], with 0..3
trailer bytes (opaque, ignored). The parser consumes 5-byte chunks
until <5 bytes remain or a freq=0 sentinel is hit. Frequency u16 BE
in Hz; gain u16 BE in **offset-binary**: raw 0..(steps-1) maps
linearly to gain_min..gain_max (so on G522 with steps=241 /
gain=[-6..6], raw=120 = 0 dB). getEQInfos returns 13 bytes with gain
bounds + step count, format enum, XY-support flag, and onboard preset
counts.
(The protocol spec lists a 2-byte header [dir_echo, slot_echo] for
getCustomEQ, but G522 firmware via the centurion bridge omits the
slot_echo and emits a 1-byte header that matches getEQDefaults.
Verified against pcap traces of LGHUB G522 LIGHTSPEED traffic.)
"""
from __future__ import annotations
@ -178,9 +182,9 @@ def parse_v2_bands(result: bytes, info: dict | None):
"""Parse a V2 getCustomEQ / getEQDefaults response.
Wire layout (see module docstring):
[direction_echo, slot_echo, band_count_max] (3-byte header)
N × [filter_type, gain_hi, gain_lo, freq_hi, freq_lo] (5 bytes)
[trailer ] (0..2 bytes, ignored)
[direction_echo] (1-byte header)
N × [freq_hi, freq_lo, filter_type, gain_hi, gain_lo] (5 bytes)
[trailer ] (0..3 bytes, ignored)
Gain is offset-binary against `info`'s gain bounds:
gain_db = gain_min + (gain_max - gain_min) * raw / (steps - 1)
@ -194,9 +198,9 @@ def parse_v2_bands(result: bytes, info: dict | None):
valid header returns []. Bands with freq=0 are treated as the
end-of-bands sentinel (matches V0/V1 behavior at lines below).
"""
if result is None or len(result) < 3:
if result is None or len(result) < 1:
return None
payload = result[3:] # skip [dir_echo, slot_echo, band_count_max]
payload = result[1:] # skip [dir_echo]
band_size = 5
if info:
gain_min = info.get("gain_min_db", -6)
@ -207,9 +211,9 @@ def parse_v2_bands(result: bytes, info: dict | None):
bands = []
for i in range(len(payload) // band_size):
e = payload[i * band_size : (i + 1) * band_size]
filter_type = e[0]
gain_raw = (e[1] << 8) | e[2]
freq_hz = (e[3] << 8) | e[4]
freq_hz = (e[0] << 8) | e[1]
filter_type = e[2]
gain_raw = (e[3] << 8) | e[4]
if freq_hz == 0:
break # disabled band — end-of-bands sentinel
if steps > 1:
@ -255,10 +259,6 @@ def get_advanced_eq_defaults(device, direction=DIRECTION_PLAYBACK, slot=0):
result.hex(),
)
return None
# Log raw=... too — getEQDefaults appears to use a different header
# framing than getCustomEQ on G522 (decoded values come out shifted
# by a byte). Capture the raw bytes so we can pin down the actual
# layout difference and adjust the parser accordingly.
logger.info(
"AdvancedParaEQ getEQDefaults V2 (dir=%d slot=%d): %d band(s) raw=%s %s",
direction,