Upsampling and specifically DXD?

On an audio forum post related to the Rossini, a question – of What is DXD – was asked. A response was: “DXD is PCM.” “Plenty of dCS owners prefer dCS doing PCM vs DSD. It’s more incisive.”

Until this I believed that DXD was some sort of upsampling method. However, when I hear PCM (Pulse Code Modulation) I don’t necessarily think upsampling, I think Redbook vs H/D vs DSD.

Suffice it to say that I am confused and would like to know more about upsampling.

I’ve read the DSD, DSDx2, DXD descriptions found in the Rossini manual (hopefully I correctly attached a pic. of them), but would like a better understanding of what exactly is going on? What can I expect to hear by changing between this 3 upsampling options?

Although until the Rossini, I own and have played many H/D files and DSD files, I never used an upsampling music player or had a DAC like the Rossini that upsampled/oversampled and am confused as to the whole premise and which option does what? I realize that I can switch between the 3 settings and hear for myself the differences. But I would still like to better understand what is actually being done to the original FLAC file.

1 Like

PCM is Pulse Code Modulation and it means that the analog signal is measured at some fixed rate – the sample frequency (e.g. 192,000 samples per second) – and each measurement has some bit depth (usually 16 or 24 bits).

CDs are 16bit / 44,100 Hz so the analog signal level was measured (or reproduced) based on 16 bit samples that were taken every 1/44,100th of a second.

DXD is a shorthand for high-rate / high-depth PCM and it corresponds to either 24 / 352,800 or 24 / 384,000. The reason for two rates is that most audio processing is done at rates that are multiples of either 44,100 or 48,000 (and the history / reasoning behind that is a long, long story).

DXD isn’t an upsampling method, it’s just a short and easy way of saying, “24 / 352,800 or 24 / 384,000.”

DSD is a different format that is based on a completely different approach to sampling theory. In it samples are taken at a very high rate – 2,800,000 Hz – but each sample only has one bit of resolution. Furthermore where each PCM sample represents the actual amplitude of the signal at that point in time, DSD is really only able to tell if the amplitude was increasing or decreasing relative to the prior sample. The very high sample rate makes it possible to reconstruct an analog signal using this method. DSD is roughly equivalent in available resolution to 24 / 176,400. DSDx2 doubles the sample rate and is effectively equivalent in available resolution to 24 / 384,000.

The three options in the Rossini mean that incoming PCM data is treated in one of three ways.

  • Upsampled to DSD (1 bit / 2,800,000 samples per second) and then processed by the Ring DAC
  • Upsampled to DSDx2 (double-rate DSD or 1 bit / 5,600,000 samples per second) and then processed by the Ring DAC
  • Upsampled to DXD (either 24 / 352,800 or 24 / 384,000 depending on the incoming sample rate) and then processed by the Ring DAC.

Incoming DSD or DSDx2 is not upsampled and is passed straight through to the Ring DAC.

The end result is the DAC itself being fed by a very high-rate stream which makes it possible for subsequent processing to be done in such a way that filtering happens well outside the audible band.

The differences are subtle and there’s no one “best” method. If there were then you wouldn’t be presented with a choice as we would have already made it for you. Furthermore, the differences are more reliant on how your brain interprets sound via your ears so it’s impossible for me to tell you what you are going to hear.

The frustrating advice here is that you need to listen to the various options and decide your own preference. Some people are very sensitive to different types of acoustic anomalies (phase, time, distortion, etc) and some systems / rooms will amplify those effects. Those people tend to be very sensitive to the various upsampling, mapping, and filter settings and in many cases those preferences change with the incoming sample rate. Other people don’t have that burden prefer a number of different combinations (and in many cases the defaults).

The short of it is that there is no way for me to tell you what you should hear nor what the best combination of settings is. It would be equivalent to me saying that green is the best color and you having a strong preference for red. There’s no reason why one or the other is better nor why there should be a preference, yet we have a strongly-rooted difference of opinion. That difference can be related back to the individual structure of our eyes, neurochemistry, our environment, and conditioning. There’s no universal right answer, but we each have a preference for what feels right.

My advice to you is to avoid trying to A / B different settings and instead listen to yourself. Start with the defaults and get used to the way it sounds. After some time (days) change the upsampling setting and then see how you react to it. Not immediately, but over time. If you find yourself enjoying it more and listening longer then you’re likely on the right path. If you’re losing interest then try a different setting and see what happens. Over time you’ll zero in on the “best” setting(s) (for you) because you’ll no longer have the desire to mess with it and instead just want to enjoy it.


A perfect response Andrew! It was spectacularly cogent and exceeded my expectations.

I was trying to comprehend what the heck was transpiring and you filled-in the blanks with aplomb.

In hindsight, I knew that the filter and mapping choices would be quite personal, extremely system and room dependent – just what my many years of experience have taught me about this crazy hobby of ours. But I do appreciate your comments and suggestions. Cheers, -Mike.


Although Ben Zwickel – the author of the subject article – has a secondary motive for its composition (the sale of his Mojo Audio, Mystique DAC) **I found the article related to my initial query and to be quite informative: **“DSD vs. PCM: Myth vs. Truth” “DSD vs. PCM: Myth vs. Truth”

1 Like

Will be interesting read @Andrew ‘s comments about this article applicable to dCS Ring DAC/filters architecture

1 Like

@Andrew I got further questions after reading your great explanation here and the different in filters you have written in Roon forum ( I am a Bartók owner and I don’t have DSDx2 setting. Based on your explanation here DSD is similar to 24/ 176.4 If my input is higher then that frequency, is that mean if the signal will somehow downsample similar to 24/176.4 before processing? Even the input is lower than that, the DSP is first upsampled to DXD, then add DSD before sending to DAC. The resolution is a bit different, how to accomplish this internally.

The other question is when the input is PCM and having DSD as upsample method, the filter using is F1 to F6, not F1DSD to F4DSD filter. So actually which set of filter is using when selected using DSD upsampling method?

Thanks a lot.

So is DSD upsampling agnostic of the music sample rate and can an external clock provide a single fixed frequency clock signal?

Not quite “agnostic”, AFAIK, dCS uses the appropriate math for the relevant incoming sample rates; 44.1kS/s based PCM streams upsampled to to 5-bit/2.822M, and 48kS/s based streams to 5bit/3.072M (and double those rates on the Vivaldi 2.0).

Anup I think that your figures relate to the oversampling which is part of the Ring DAC process. The 5 bit is the giveaway.

I believe that that OP is referring to use of a dCS upsampler where a 48KS/s integer source can indeed be upsampled to DSD . However his question is about clock frequency and he has been referring to his use of a Mutec clock in his other posts to which this question clearly also relates.

I cannot even attempt an answer in this context as we have one correspondent saying that the Mutec device switches frequency automatically whereas another says this is only available when the Mutec is reclocking data ( another thing altogether) and the user manual ( given my skim reading of it) implies that it is only available via its USB output.

I go along with an answer to his other Mutec question.If he is going to buy one of the current generation of dCS DACS then he should buy the matching clock. He will then be certain that all will work as intended rather than risking problems brought about by introducing an “unknown”. We do at least know that the Mutec has no double independent clock outputs which the current DACs really require.

Whoops. Ok, Guess I’m lost on the context as well!

ps: :blush: I only just read the past posts on this thread, there’s tones of info from Andrew!

Can any 1 plz explain the difference they hear when you switch dxd to dsd to dsd2? What you prefer and why. Just your personal experience

This is an old but very helpful explanation but one input possibility has been left unanswered and I just want to make sure.

Original statement by @Andrew:
“Incoming DSD or DSDx2 is not upsampled and is passed straight through to the Ring DAC.”

Is this also true for incoming DXD? That it is not upsampled?

Thank you!

As DXD already uses the maximum PCM sample rate that current dCS converters are able to process what could incoming DXD be upsampled to as it would lie beyond the device’s capabilities?

Sorry for the dumb question, just trying to understand.

Apology not necessary, not a dumb question :smiley:

All good Pete.
The news is so bad these days I’ve decided to direct my free time to reading old dCS posts and coming up to speed on random esoteric topics (!)

I’m also considering a substantial system upgrade including a pair of Ktemas, as I have deep admiration for Franco Serblin…

I would say two things primarily are changing: increasing impact from clock jitter (audible) and increased ultrasonic noise (not audible).