There are multiple noise floors that matter here. The graph shows noise from nonlinear issues with sigma delta modulation.
That graph was produced by a averaging multiple FFTs each of which has many points (to get the desired frquency resolution.) If a million point FFT is used it will average white noise over a million bins and hence lower the apparent white noise power by 60dB. Also since multiple FFTs are averaged, white noise is also lowered by the square root of the number of FFTs that were averaged.
The purpose of these graphs is to make sure that non-white noise is clearly shown, but to get an idea about the white noise you need to know how many points were used in the FFTs and how many separate FFTs were averaged.
Double rate DSD allows moving the knee of the rise of nonwhite noise up an octave, quad rate another octave… BUT, higher sample rates are also compromised by more jitter which adds more “ugly” noise everywhere. So there is a sweet spot that depends somewhat on the other parameters of your implementation.
There are other reasons to like double or quad rate for upsampling - the more frequency room you have the more options for tailoring the noise floor or other tradeoffs. With single rate DSD the options are fairly constrained.
Not at all! I think the Format thing is no longer terribly relevant in most applications, unless chasing whatever the Ultimate Current Thing is matters to you for some reason. I’m usually blissfully ignorant of what the sample and bit rate is of whatever I’m listening to. It has been a decades-long road during which that was a lot more relevant, and so it is a big relief.
That is correct. Sonoma is only capable of DSD64 (or 1XDSD). We are working on developing a new system that can handle 4XDSD but that’s going to take some time. Some companies upsample but that’s not the pure approach we would want to take.
Sounds like things are getting closer to recording in DSD256 rates , from your latest video. Much much better than the old outdated Sonoma system at DSD64 which has lots of technical issues, as per earlier posts.
Merging and RME have had DSD256 recorders for a while now.
I assume this means DSD256 recordings will start to be available soon too?
Roughly the sweet spot is different for each bit width, i.e. it’s different for one bit than, say five bit or sixteen bit. One bit is the most sensitive to jitter so it’s lower for one bit than other sample widths. It’s the same for the DS Mk II and the TSS as the DS. At the next level of detail I can’t really say whether lower jitter of the Mk II and TSS will do more or less to raise the sweet spot than their lower noise.