I guess it’s my turn to say that your opinion about my opinion is just your opinion. I’d appreciate it if we ended the opinion game here.
DACs don’t upsample they oversample and that is not the same as upsampling, similar yes, same no.
To achieve oversampling, DACs need to re-sample at a higher rate an already-sampled signal. That is up-sampling. Oversampling is not up-sampling only when the analog signal is sampled directly at the higher rate. ADCs do that, not DACs.
Sorry but t’s done in DACs to at the sigma Delta phase NoS stands for non over sampling not non up sampling or it would be NuS not NoS.
I am not the one telling people that what they want is beyond the limits of which they can make use or that it is just marketing. I am not the one stating opinion as fact.
1 Like
I guess the term “oversampling” is used to denote a process that obtains a signal sampled above the Nyquist limit, and it is used both in ADCs and DACs. In DACs, it is achieved by up-sampling.
1 Like
@Mikael_Ollars , it is already a challenge for most processors to handle DSD512, including the NUC 10th Gen i7 and the Nucleus+. What percent of Roon subscribers would possible invest in new hardware to play DSD1024? I suggest it would be an insignificant number.
I also suspect Roon has to prioritize their enhancements based on actual demand that will increase their subscriber base. Again, I doubt this will add the next 100,000 subscribers.
You are correct; GoldenSound is free to ask for any enhancement s/he likes. Best of luck to him/her.
1 Like
this is going to be fun
2 Likes
just because you don’t use or want to use a feature does not mean that others shouldn’t be allowed to use it if they want. how others enjoy music should solely be their decision.
3 Likes
Fact: More bits, more taps, higher frequency range, the more accurate the maths are in the areas that were explained by @GoldenSound above.
Fact: More bits, more taps, higher frequency range, you get a different and more precise signal out the analog side of the DAC. Again, examples above.
Not Fact: You can hear a difference from the two facts above. 
I’m strongly of the opinion that I don’t care if I or you can hear it. Higher rates, PCM or DSD, produces a better measurable result so let those with equipment to measure go forth and measure. I like this feature request and I voted for it. There is plenty of gear out there that can take advantage of it. If I get the chance to obtain such gear I’ll certain be one to try it as well.
5 Likes
They are the same thing. Or rather, there is no particular set definition splitting the two.
Oversampling/upsampling are alternative words for the same mathematical process.
Typically you’d see ‘oversampling’ used more in live-playback situations like in a DAC and ‘upsampling’ used more in static scenarios where the data is being saved to a file not played back, but there is no set definition.
2 Likes
Wholly agree, but that is not what being requested here. Golden Sound is requesting Roon to identify and add media files in those extreme resolutions, for a bit perfect transport to the compatible DAC.
So, not upsampling in real time, rather recognition and management of these HUGE files!
6 Likes
I actually think 1.536mhz or DSD1024 should work in Roon, regardless of insanity - Roon should reflect the support in the DAC 
9 Likes
Since up-sampling can’t make up data that is not there, how can you possibly achieve a higher frequency range? That is locked in at the time of A/D.
2 Likes
If you don’t include a short explanation of what you’re trying to prove with your link, you might as well link to this:
4 Likes
Sure, others can explain this better (or correct me) but from my limited understanding…
At 44.1 any errors end-up back in the audible frequencies because that’s the only place for them to go. When you increase the frequency range, up-sample / oversample, the errors land outside the audible range leaving the audible portion of the range more precise and cleaner (free of errors / unwanted harmonics).
It’s not about adding information. It has to do with giving enough resolution to the filters and modulators to not damage what’s in the audible range. The link I provided gives an example of what a clipped signal sounds like at various oversampling rates. You can hear a difference, not because you’re hearing the higher frequencies of the oversampled signal, but because the higher oversampling rates reduce the errors / harmonics being pushed down into the audible range.
The point is, your DAC oversamples already. All DACs, except a true NOS DAC, oversample for this reason of reducing errors in the audible range. The argument, and this can be proven with the maths, is that the higher frequency you can use to oversample then the less errors arrive within the audible range. So, why don’t we all go to this extreme resolution always? Because it’s expensive…
Either you pre-process the data which creates massive files and potentially a storage problem. You do it in real-time in the DAC which significantly increases the DAC cost (think Dave + m-scaler and that’s achieving 768kHz resolution). Or, you do it with something like HQP in realtime which, currently, requires some pretty beefy hardware but technology moves on and what’s hard today becomes a blip in utilization on your watch a couple years from now.
3 Likes
Thanks, I appreciate it!
I believe you are talking about quantization errors. Yes, if you sample at 44.1 kHz, AND if you use 16-bit samples, quantization errors have to be spread more or less throughout the audible range. (For samples with 24 bits or higher, quantization errors are completely negligible, as the theoretical noise floor, even in the absence of dither, is no greater than -144 dB). But once you have an audio file, sampling and quantization have already happened, and there is nothing you can do at D/A time in terms of DSP to undo their artifacts. What you CAN do - and should do - is to make sure not to introduce additional in-band errors during D/A. This is the scope of the problem, right?
Now, when you up-sample (or over-sample if you wish) to 128fs or 256fs, you add 7 or 8 octaves above the audible range, respectively. That extra bandwidth is so large, you can get away with using only one bit samples (i.e. DSD). One bit samples generate the highest quantization errors possible. Yet, you can practically push ALL quantization noise really high up the frequency range, far enough from the audible range so that you can use a VERY relaxed analog filter after the D/A conversion stage.
What would the point be of adding 2 or 3 extra octaves to the bandwidth with DSD1024, when you already have plenty? For the purpose of further relaxing an analog filter that is already relaxed, you add a significant burden on the D/A stage, which now has to cleanly switch at 4x or 8x higher frequencies. Nothing to gain, much to lose.
Also, you don’t need higher up-sampling ratios to use more taps. You can use 5B taps for a 2x ratio just fine by simply increasing the size of the time window (i.e. by taking more and more adjacent samples into computation). There is no need to couple the number of taps to the ratio. This means that the accuracy of the math is not dependent on the ratio either. In fact, given the same number of taps, the accuracy decreases as the ratio increases.
Why are so many people in this thead so entirely unpleasent? If you dont care about the feature dont vote for it and move on instead of being miserable
20 Likes
HQplayer is the way to go. For rates that high you need really good filters to remove all the noise created by upsampling. HQplayer has the best filters for this.
5 Likes