Monorepo for Aesthetic.Computer
aesthetic.computer
native: bump dither keepalive to ±160 LSB (±1 extended to 96s, not enough)
±1 LSB extended sdmode-on from 23s to 96s but SOF's silence detector
uses RMS threshold, not zero-detect. ±160 LSB is -44 dBFS (~0.6% FS),
still inaudible at normal speaker levels, but well above the DSP's
silence gate. Should keep the speaker amp permanently powered.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
+13
-9
+13
-9
fedac/native/src/audio.c
+13
-9
fedac/native/src/audio.c
···
1489
1489
buffer[i * 2] = (int16_t)(mix_l * 26000);
1490
1490
buffer[i * 2 + 1] = (int16_t)(mix_r * 26000);
1491
1491
1492
-
/* DAPM keepalive: inject ±1 LSB dither when the buffer
1493
-
* would otherwise be pure zeros. SOF DSP's internal
1494
-
* silence detector powers down the SSP1 BE DAI after
1495
-
* ~20-40s of all-zero frames, which drops MAX98360A
1496
-
* sdmode to 0 (speaker amp off). ±1 at 16-bit S16_LE
1497
-
* is -96 dBFS — utterly inaudible, but keeps the
1498
-
* pipeline alive and the amp powered. */
1492
+
/* DAPM keepalive: inject low-level dither when the buffer
1493
+
* would otherwise be silent. SOF DSP's internal silence
1494
+
* detector uses an RMS threshold (not just zero-detect)
1495
+
* and powers down the SSP1 BE DAI, dropping MAX98360A
1496
+
* sdmode to 0. ±1 LSB extended amp-on from 23s to 96s
1497
+
* but wasn't enough. ±160 LSB is -44 dBFS — still well
1498
+
* below audibility at normal speaker levels (~0.6% of
1499
+
* full scale) but gives the PGA enough RMS energy to
1500
+
* keep the pipeline active. The alternating pattern
1501
+
* averages to DC=0 so there's no pop/click risk. */
1499
1502
if (buffer[i * 2] == 0 && buffer[i * 2 + 1] == 0) {
1500
-
buffer[i * 2] = (i & 1) ? 1 : -1;
1501
-
buffer[i * 2 + 1] = (i & 1) ? -1 : 1;
1503
+
int16_t d = (i & 1) ? 160 : -160;
1504
+
buffer[i * 2] = d;
1505
+
buffer[i * 2 + 1] = -d;
1502
1506
}
1503
1507
1504
1508
// HDMI audio: 1-pole low-pass filter + downsample