OK, so I gave it a second thought - and more measurements (all with 15.8 Ohm load)
1. Measured Noise is actually slightly better for 40mV in M gain than in L gain (attenuation through the volume knob)
We speak 0.5dB, so nothing significant. But that's the fact.
2. Dynamic range - Un-weighted - AES 17 method (so with a -60dB 997Hz signal 20Hz-20kHz, volume knob at max), we get
For Med and Low, that's noise at 1.44uVrms and 0.78uVrms respectively
So the <1.5uVrms and <1uVrms are correct (especially A-weighted, like in the specs, since my measurements are un-weighted)
3. Analysis
So, given measured Max level at 0dBFS, I can compute the at.tenuation for 40mV
If I deduct it from the measured DR figure, that gives us an expected SNR at that level.
The Low gain should have a better SNR than Medium gain. We speak several dBs.
I measured with digital attenuation (lowering the level from the measurement software, with volume knob at max) and I get SNR figures pretty close to expected (0.18dB lower, which is not relevant).
Then, that doesn't match with measured SNR when DAC is fed by a full scale signal and level is attenuated with the volume knob.
We see higher figures than with digital attenuation for both gains, and Medium gain actually outperforms slightly Low gain.
So the bottom line is:
The volume knob (analog) attenuation is more efficient than the digital attenuation by a margin of several dBs.
This better efficiency allows to keep more of the higher dynamic range of the Medium gain, which gives a small (but non significant) edge, in your use case, to Medium gain over Lower gain.
Even if the later, at same DAC-output level, has less noise.
That was fun .
Thanks.
