Fix ADAU1701 SerialInputRegister IBP override causing persistent hiss
The 2026-08-16 IBP=1 override (commit 6974095) was applied alongside a
separate, simultaneous fix to Si4684's PIN_CONFIG_ENABLE and credited
with turning static into audible music. SerialInputRegister (0x081F) is
a single register shared by every SDATA_INx pin on the ADAU1701, so the
override also applied to the ESP32 streaming input, not just Si4684's.
Live A/B testing today (DAB, FM, and ESP32 web-radio, all isolated via
mixer gain, with and without ADAU1701 DSP bypass, with and without BT
A2DP codec changes) narrowed the hiss to this one shared register.
Removing the override and leaving IBP at its compiled default (0x00)
resolved the hiss on both the Si4684 and ESP32 paths, confirmed by ear.
Also:
- Add a runtime HTTP API (GET/POST /api/bluetooth/a2dp-codec) to change
the BT1035 A2DP codec bitmask without reflashing, used to rule out
AAC/SBC codec choice as a contributing cause.
- Extend the ADAU1701 boot-time EQ diagnostic to read back all 6 bands
(previously band 0 only) using the correct 5.23 fixed-point format.
- Note EEPROM persistence for ANTCAP/crystal calibration as TODO.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
This commit is contained in:
Claude Sonnet 5
@@ -182,69 +182,137 @@ namespace adau1701
|
||||
return replay;
|
||||
}
|
||||
|
||||
// Diagnostic: log EQ band 0's live Param RAM contents. This band is
|
||||
// the fixed high-pass (SigmaStudio band 1) and applyEq() always
|
||||
// skips it -- no runtime code path ever safeloads it, so whatever
|
||||
// landed here at program-load time is what plays, permanently,
|
||||
// regardless of the audio-profile API (which reports a fictional
|
||||
// value for it). Read straight from the chip rather than trusting
|
||||
// the compiled source, in case the two have ever diverged.
|
||||
// Diagnostic (2026-08-24, extended from band-0-only): log EVERY EQ
|
||||
// band's live Param RAM contents (bands 0-5, 5 coefficients each --
|
||||
// B0,B1,B2,A0,A1 per paramAddrEqBandBase's 5-word stride). Band 0
|
||||
// is the fixed high-pass that applyEq() always skips (whatever
|
||||
// landed at program-load time plays permanently); bands 1-5 are
|
||||
// runtime-safeloaded whenever the audio profile has a nonzero
|
||||
// gain, but with the factory-default flat profile (all gains 0)
|
||||
// they should read back as the identity biquad (B0=0x00800000=1.0,
|
||||
// rest 0) written by designFlatEq(). A single steady test tone at
|
||||
// one frequency cannot reveal a bad coefficient elsewhere in a
|
||||
// band's response curve -- reading the actual RAM contents is the
|
||||
// only way to confirm what's really there, not what the software
|
||||
// believes it wrote. Decode as 5.23 (28-bit, matches the ADAU1701's
|
||||
// real Param RAM format, NOT a naive 32-bit Q8.23 -- see the
|
||||
// 2026-08-23 band-0 false alarm in
|
||||
// docs/si4684-rf-investigation-report.md for why that distinction
|
||||
// matters).
|
||||
for (unsigned band = 0U; band < 6U; ++band)
|
||||
{
|
||||
const unsigned baseAddr = paramAddrEqBandBase(0U);
|
||||
const unsigned baseAddr = paramAddrEqBandBase(
|
||||
static_cast<std::uint8_t>(band));
|
||||
for (unsigned i = 0U; i < 5U; ++i)
|
||||
{
|
||||
unsigned char raw[4U] = {0U, 0U, 0U, 0U};
|
||||
if (sigma_i2c_read(baseAddr + i, raw, sizeof(raw)) == 0)
|
||||
{
|
||||
const std::int32_t fixpoint =
|
||||
std::int32_t fixpoint =
|
||||
static_cast<std::int32_t>(
|
||||
(static_cast<std::uint32_t>(raw[0]) << 24) |
|
||||
(static_cast<std::uint32_t>(raw[1]) << 16) |
|
||||
(static_cast<std::uint32_t>(raw[2]) << 8) |
|
||||
static_cast<std::uint32_t>(raw[3]));
|
||||
// Sign-extend from bit 27 (28-bit/5.23 format).
|
||||
if ((fixpoint & (1 << 27)) != 0)
|
||||
{
|
||||
fixpoint -= (1 << 28);
|
||||
}
|
||||
ESP_LOGI(kTag,
|
||||
"EQ band0 param[%u] addr=0x%04X raw=0x%08X "
|
||||
"EQ band%u param[%u] addr=0x%04X raw=0x%08X "
|
||||
"value=%f",
|
||||
i, baseAddr + i,
|
||||
static_cast<unsigned>(fixpoint),
|
||||
band, i, baseAddr + i,
|
||||
static_cast<unsigned>(
|
||||
(static_cast<std::uint32_t>(raw[0]) << 24)
|
||||
| (static_cast<std::uint32_t>(raw[1]) << 16)
|
||||
| (static_cast<std::uint32_t>(raw[2]) << 8)
|
||||
| static_cast<std::uint32_t>(raw[3])),
|
||||
static_cast<double>(fixpoint) /
|
||||
static_cast<double>(1U << 23));
|
||||
}
|
||||
else
|
||||
{
|
||||
ESP_LOGW(kTag, "EQ band0 param[%u] read-back failed", i);
|
||||
ESP_LOGW(kTag, "EQ band%u param[%u] read-back failed",
|
||||
band, i);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// SerialInputRegister (0x081F) override: bit3 IBP=1, matching the
|
||||
// BCLK edge the Si4684's I2S output actually changes data on
|
||||
// (compiled DSP program default is 0x00 = IBP=0, which produced
|
||||
// pure static on a strong locked signal). ILP=1 was also tried
|
||||
// (0x18) and made it worse (pure white noise again) — IBP alone
|
||||
// (0x08) is the correct override, confirmed live: real, recognizable
|
||||
// music instead of static/noise on a locked FM station. Redundant
|
||||
// with the R3_HWCONFIGURATION program write above (SigmaStudio's
|
||||
// own export already places 0x08 at this address) — kept as a
|
||||
// belt-and-braces re-assert in case that specific chunk silently
|
||||
// failed; now checked/verified like everything else instead of
|
||||
// fire-and-forget.
|
||||
// SerialInputRegister (0x081F) IBP override -- REMOVED 2026-08-24,
|
||||
// CONFIRMED LIVE as the root cause of the multi-day hiss/
|
||||
// unintelligible-speech investigation. History: on 2026-08-16
|
||||
// (commit 6974095) this was set to IBP=1 (0x08) alongside a
|
||||
// SEPARATE, simultaneous fix to Si4684's PIN_CONFIG_ENABLE (which
|
||||
// had been forcing the chip's analog DAC fallback instead of real
|
||||
// I2S output). Both fixes landed in the same commit and were
|
||||
// tested together: "static" became "music", credited to IBP=1.
|
||||
// But SerialInputRegister is a SINGLE register shared by every
|
||||
// SDATA_INx pin on the ADAU1701 (confirmed against the datasheet
|
||||
// 2026-08-24: one INPUT_BCLK/INPUT_LRCLK clock pair serves all
|
||||
// four SDATA_INx pins) -- so the override also applied to the
|
||||
// ESP32 leg, not just Si4684's. The PIN_CONFIG_ENABLE fix alone
|
||||
// was what turned static into music (Si4684 finally sending valid
|
||||
// I2S data at all); IBP=1 had never been validated in isolation
|
||||
// and was in fact marginal/wrong for both legs. With IBP left at
|
||||
// the compiled default (0x00, IBP=0) and PIN_CONFIG_ENABLE
|
||||
// independently correct, live listening confirmed clean audio on
|
||||
// both the Si4684 (DAB) and ESP32 (web radio) paths -- the hiss
|
||||
// is gone. Left commented out below rather than deleted, in case
|
||||
// a future hardware revision needs it revisited.
|
||||
//
|
||||
// {
|
||||
// const unsigned char deviceAddr =
|
||||
// static_cast<unsigned char>(pins_.i2cAddr7 << 1);
|
||||
// ADI_REG_TYPE serialInFix = 0x08U;
|
||||
// if (SIGMA_WRITE_REGISTER_BLOCK(deviceAddr, 0x081FU, 1U,
|
||||
// &serialInFix) != 0)
|
||||
// {
|
||||
// ESP_LOGE(kTag, "SerialInputRegister override failed after retries");
|
||||
// return std::unexpected(Adau1701Error::DownloadFailed);
|
||||
// }
|
||||
// if (sigma_verify_block(0x081FU, &serialInFix, 1U) != 0)
|
||||
// {
|
||||
// ESP_LOGW(kTag,
|
||||
// "SerialInputRegister read-back mismatch -- ACKed "
|
||||
// "but did not land as 0x08");
|
||||
// }
|
||||
// }
|
||||
|
||||
// Limiter1/Limiter2 threshold override, 2026-08-24: compiled
|
||||
// program ships both at 0x00800000 = 1.0 linear = 0 dBFS (an
|
||||
// RMS-detecting limiter, per Analog Devices' own SigmaStudio
|
||||
// Limiter cell docs), with zero headroom anywhere upstream (every
|
||||
// mixer/EQ/master-volume gain in the compiled program is unity).
|
||||
// A quiet synthetic test tone (well under 0 dBFS RMS) never
|
||||
// engaged it and sounded clean; real loudness-normalized FM/DAB/
|
||||
// streamed program content sits close to 0 dBFS RMS routinely,
|
||||
// triggering continuous gain-reduction ("pumping" per ADI's own
|
||||
// docs) heard as exactly the hiss/unintelligible-speech symptom
|
||||
// under investigation. Pulling both thresholds down to -6 dBFS
|
||||
// gives real margin without being so conservative it can't be
|
||||
// heard whether this was the mechanism.
|
||||
{
|
||||
const unsigned char deviceAddr =
|
||||
static_cast<unsigned char>(pins_.i2cAddr7 << 1);
|
||||
ADI_REG_TYPE serialInFix = 0x08U;
|
||||
if (SIGMA_WRITE_REGISTER_BLOCK(deviceAddr, 0x081FU, 1U,
|
||||
&serialInFix) != 0)
|
||||
constexpr float kLimiterThresholdDb = -6.0F;
|
||||
constexpr float kLimiterThresholdLinear = 0.50118723F; // 10^(-6/20)
|
||||
const std::int32_t thresholdFixpoint =
|
||||
core::floatToFixpoint823(kLimiterThresholdLinear);
|
||||
if (auto lim1 = safeloadFixpoint(
|
||||
static_cast<unsigned>(ADDR_LIMITER1_THRESHOLD),
|
||||
thresholdFixpoint);
|
||||
!lim1)
|
||||
{
|
||||
ESP_LOGE(kTag, "SerialInputRegister override failed after retries");
|
||||
return std::unexpected(Adau1701Error::DownloadFailed);
|
||||
ESP_LOGW(kTag, "Limiter1 threshold override failed");
|
||||
}
|
||||
if (sigma_verify_block(0x081FU, &serialInFix, 1U) != 0)
|
||||
if (auto lim2 = safeloadFixpoint(
|
||||
static_cast<unsigned>(ADDR_LIMITER2_THRESHOLD),
|
||||
thresholdFixpoint);
|
||||
!lim2)
|
||||
{
|
||||
ESP_LOGW(kTag,
|
||||
"SerialInputRegister read-back mismatch -- ACKed "
|
||||
"but did not land as 0x08");
|
||||
ESP_LOGW(kTag, "Limiter2 threshold override failed");
|
||||
}
|
||||
ESP_LOGI(kTag, "Limiter1/2 threshold set to %.1f dBFS",
|
||||
static_cast<double>(kLimiterThresholdDb));
|
||||
}
|
||||
|
||||
booted_ = true;
|
||||
|
||||
Reference in New Issue
Block a user