Hi Mani,
The answer is in this quote from the Release Notes on 0.9u-1 :
- When "DAC needs" is set to 32 (!!!), hence the "DAC is" > 16 bits, Doubling and Upsampling (to 88K2) is supported now.
Important : Most often Upsampling takes place in the 16 bit domain, as was the case with XXHighEnd before. Now, "genuin" Upsampling is performed as soon as the DAC has more than 16 bits (17 would be sufficient theoretically). BUT :
Upsampling *and* using the Digital Volume Control should really not be applied together !!
Why ? because the "genuin", or better, "balanced" upsampling will be destroyed by it. Anyway in my theories the Upsampling of the Samplerate (2x) including interpolation to the proper amplitude values (right in between two adjacent original 44K1 samples' amplitude) will be destroyed by the Digital Volume. And just the same : when an e.g. 18 bit DAC allows for the headroom to do this, attenuating with 12dB already has removed that necessary headroom.
When your DAC is 24 bits and play it so ("DAC needs" is set to 24 (but see below topic for 24 !) or 32), YMMV and it kinds of depends how much you attennuate.
On the other hand : When you use your system with the Digital Volume set to -0dB (hence the Volume Control is analogue), Upsampling will be way better than before hence will not be able to show the anomalies from before, thinking of no headroom at all when the amplitude varies less than 2 decimal in two adjacent samples, or varies uneven decimal for that matter (like 3, 5, 7, etc.).
So ... this is an indirect implication of using the additionally emerged space between two adjacent samples in the
bit depth domain. It appears to be difficult to reason out towards you, but luckily it was predicted (see quote) and is not something I can't understand.
Thinking in "headroom" terms as per the quote, will at least give you the idea that this has to come from somewhere, and obviously the one bit needed for that can only emerge at ... the lower (Least Siginificant Bit) side. Technically it comes down to dividing decimal 1 by 2, which sets the
originally available LSB to 0, and uses up the one (available at > 16 bits DACs) below that, and set it to 1. Decimal this would come down to 0.5 (where the base for smallest value WAS 1).
When 4 x Upsample is available within XX (to 176400), you will see that 2 additional bits will be consumed at the LSB side.
As per your pictures, people will be able to better understand what happens (and which was so beautifully "found" by Russ earlier) : in your case you have 7 bits (42dB) left to attennuate this Upsampled situation. Me, with my 18 bits DAC would have 1 bit (6dB) left. With a 16 bit DAC there's no headroom to start with ...
Btw, I never realized that I could have shown/explained this by means of DigiCheck, which I just have the same (coming along with the RME Fireface).
Peter
