Cirrus Audio chip linearity error above -35dB

[TheTalbotHound]

Does anyone know what causes the slight wobble in linearity from Cirrus chips above -35dBfs?

With the apple dongle it seems to manifest a bit differently.

 

[Killingbeans]

At first I was like "what error?"

Looks like it's more like a zero crossing at -20dBfs, and then it rises a tad on the lower levels and dips a tad on the higher ones.

It's like +/- 0.05dB, maybe +/-0.1dB max? Nothing even remotely of audible concern? Still a funny little quirk.

 

[TheTalbotHound]

At first I was like "what error?"

Looks like it's more like a zero crossing at -20dBfs, and then it rises a tad on the lower levels and dips a tad on the higher ones.

It's like +/- 0.05dB, maybe +/-0.1dB max? Nothing even remotely of audible concern? Still a funny little quirk.

Yeah maybe not audible, but i still want to know how or why something like this could come about. Especially since it is so consistent across Cirrus chip devices.

 

[AnalogSteph]

This quirk seems to affect all the modern tiny mobile chips and their relatives, i.e. CS4399, CS43130, CS43198 and CS43131. I would like to know, too. It is lower than I would expect if it had something to do with the Class G output stage, at the same time a thermometer DAC based concept has no inherent kind of linearity that could cause this so it's all a bit of a puzzle.

The only thing I could think of is that DAC reference voltage might be sagging a bit once analog supply sees some load (I mean, you wouldn't want to be implementing any more voltage regulators than necessary, and a shared +1.8 V might be our culprit), but even then you'd be expecting most of the error up top, not this sort of transition between -12 and -32 dBFS.