Schiit Yggdrasil -- 21 bit?

@gdhal

I don’t want to argue either. I posed a simple question in the OP. I got some good answers, but was also viciously attacked by some aggressive individuals here.

One’s logic almost compels one to ask: how can you get 21 bit resolution out of 20 bit DAC chips. Since independent verification is essential in science and engineering, JA’s (and other people’s) measurements are crucial. That is why we have reviews.

You are not against independent reviews, are you?

You can still get some signal resolution above 20 bits by dithering the 24 bit digital signal when converting to 20 bit. Dithering is much preferred to just truncating the least significant bits. Dithering adds random noise to the signal and raises the S/N but it will preserve some signal below the LSB. Studios do this all the time when taking 24 bit Masters and producing a 16 bit CD. 

I think Schiit are using the same logic as Meridan in saying that the last 3 bits on 24 bits is just noise (below current analog dynamic range resolution which is about 21 bit equivalent). This is true on a full scale signal. However a good recording may not use all full scale 24 bits so the last 4 bits could very well be within analog performance range and it is bad design to say we just throw them away. Also dithering is a form of processing and compression and just like with lossy audio and photo compression - some algorithms are bettter then others. 

Do do you trust your DAC manufacturer to process and massage the incoming 24 bit audio and throw away 4 bits or do you prefer to hear what the studio mastering engineer carefully produced?

How can the DAC be 21 bit if the chips are 20 bit?

The two DAC chips on each channel are described as being used in a balanced configuration. So I would suspect that one chip handles one half of the eventual output voltage range and the other chip handles the other half. Which would mean that 20 bits of resolution is provided for each half of the output voltage range, which in turn would correspond to 21 bits of resolution relative to the entire range.

That would be the theory, at least. How well it works out in practice will depend on the overall design, of course.

Regards,
-- Al

@almarg 

Great and very clear answer. Many thanks, super helpful.

You are not against independent reviews, are you?

Of course not.

Your original question is valid, however, it hadn't really been answered (accurately or inaccurately) until after my post, when shadorne provided additional clarification and almarg offered yet another possible explanation.