Well, re-reading this thread (or at least the key posts), I have to confess that my understanding of 518 resolution enhancement - specifically the word length aspects of resolution enhancement - was mistaken. In part, I think this is because the otherwise excellent 518 manual doesn't explain how DAC word-length matching is actually achieved.

I had assumed that the top 16-bits would be stretched to 24-bits, filling the (previously zero) least significant 8-bits with extrapolated signal. This is wrong and, as I now realise from George's excellent posts and diagram, the top 16-bits must remain where they are.

However, I didn't understand George's diagram, until I read this very good paper by Wadia. When I did, the penny dropped and everything fell into place.

http://www.wadia.com/technology/technicalpapers/Resolution_Enhancement.pdf

The Wadia paper is entirely consistent with what Bob Stuart has written on resolution enhancement in the 518 manual (page 49ff); that section now makes a lot more sense to me, particularly how the different dithers are applied and how the benefits of resolution enhancement to the DAC - namely increasing effective DAC resolution and linearity - are achieved. In short, the bottom 8-bits are filled with jitter de-correlating dither.

As an aside, I'm not sure whether digital gain is applied to the top-most bits (I.e. the original signal) before adding the LSBs, or to the 24-bit signal afterwards. It doesn't really matter though.

Roon Developer and ex-moderator of this Forum
I am #25 in the HH1 photo of fame.