Output stage
The last stage before the DAC. By default every part of it measures or counts, and none of it alters the signal.
Bit depth and dither
In float there is nothing to dither.
Dither exists to decorrelate quantisation error from the signal. While the output is floating point, no quantisation happens on our side, so applying it would only add noise.
macOS always receives float from us.
Even a device that reports an integer format is handed float32, and the system converts at the very end. That conversion is exact only when the values already sit on a grid that float32 represents exactly, which is why the depth setting matters even though the handover is float.
TPDF is the only dither offered.
Noise shaping is not, because it moves energy into a band rather than removing it, and at these rates the decision belongs to your own design rather than to a fixed output stage.
True peak and auto headroom
true peak: 4× oversampled, to BS.1770
ceiling: -1.0 dBTP
release: 200 ms, first order
off nothing happens
detect over-ceiling samples are counted. The signal is not changed (default)
auto the gain is lowered by exactly what it takes to stay under, then released- The default counts rather than acts, because changing the signal by default would contradict what the rest of the chain is for.
- A graph containing a nonlinear node cannot start without the limiter, and it stays armed for the whole session. That is the one case where the output stage acts whether or not you asked.
- The amount of gain reduction being applied is always shown, so a quiet passage is never quietly explained by the output stage.