The sound of DSD

I have just noticed this on the HQPlayer main screen. Does this provide a warning that I should be using an apodizing filter or is it used for something else?

Yes, if the apod counter increases to values over 10 or so, you should be using an apodizing filter.

3 Likes

Guess it’s time for me to respond. Was very busy this week and was ingesting all the great info / discussion before I said something foolish… which I might do anyway :stuck_out_tongue:

I’ll start by thanking @andybob for starting the thread. I think you captured my comments well. But, I also think you and I realized our opinions were formed long ago and we’ve not done any “modern” comparisons. The horrible thing about time is it ages us and… well… that comes with changes in hearing which is why I should probably drop old opinion and start over. The timing is good though… I am looking to upgrade my DAC late this year maybe and discussions like this are driving my decision.

To answer @dabassgoesboomboom
I don’t currently have a DSD DAC. My DSD opinion was formed when I did a lot of SACD listening many years ago and compared to PCM via the same playback (SACD player acting as DAC). Maybe that skewed my opinion… maybe I need a reset.

I do have a bias for DSD and, after some various other upgrades on my list, my goal is to implement HQP sending DSD256 or DSD512 to something like a May. At that point I’ll be able to run some A/B testing between formats at rates where the returns should be very minimal. Maybe then I’ll lose my bias. Not being a gamer all my current PC hardware is unable to achive this which means I’ll be buying DAC plus compute at the same time. I’m trying to match the DAC to the greatest benefit of having HQP and budgeting accordingly.

@jussi_laako Much appreciate the clarification there. I was trying to extrapolate a change in a set of bits to voltage vs. deriving a change in voltage from swapping between 1’s and 0’s over time and… well, consider me an Internet hobbyist with my level of understanding. Where I just completely missed the mark, and I realize that now, is that the voltage change derived from PCM does require multiple samples. Your comment about NOS DACs is interesting. I’ve never considered a NOS DAC without upsampling. The more I read the better I understand the benefits to such a pairing.

Would it be safe to say that the voltage change, regardless of LPCM or SDM source, becomes more accurate with more samples? Which, in my mind, gives positive benefit to high resolution regardless of if the high resolution is via source file or upsampled? This, ultimately seems to be the core benefit of both HQP and a product like M-Scaler.

I’ll make some very generic conclusions going back to my original comment on sound that I find DSD to be softer, more analog, slower and PCM leaving a harsher, but more defined, “edge” in transients. From reading this discussion, with enough samples and proper filters, PCM and DSD should sound the same. There isn’t any reason for them to sound different once enough samples are provided to reduce errors in the DAC. Therefore, the changes being heard are based on complexity of a) obtaining enough samples b) the complexity of filters needed for PCM vs DSD within the DAC.

As an Internet hobbyist… I’ll conclude

  • It’s easier to get “accurate” feeding DSD to a DAC vs. PCM regardless when feeding similar number of samples between the two.
  • Both 16/44.1 and DSD64 can, technically, capture an accurate 2-channel waveform within the spectrum of human hearing but neither are really great on the playback side for reproducing that captured waveform.
  • Hence, we should be upsampling somewhere or seeking out high resolution sources to give the DAC a better chance at accuracy (upsampling in the DAC is my current set-up).
  • As with everything in audio… listen to what brings you joy over everything else and don’t worry so much about the technical aspects of “why” :slight_smile:
2 Likes

I have a chord dac and pcm I found preferable. Dsd gets converted to pcm so there is an unnecessary conversion. Dsd has a clarity edge at expense of less fullness. So dsd in roon gets converted to pcm in the dac regardless, so pcm to pcm may just be more musical? I don’t know about dsd dac players like pure stream.

how do I get access to listen and participate on this weekly zoom call?

1 Like

See you Saturday:

New link / bridge posted each week so watch that link and @Rugby will post the link once its open.

:grinning:

A bit more on the recording side as they are recording in DSD64 with Sonoma system at their Octave Records studio. They are planning to make changes in future to record in DSD128.

1 Like

I have a few Octave Recordings. They do sound good if a bit dry. Although, I’m converting them to PCM until I get my DSD playback chain in place. I do hear very slight differences between the PCM files and the DSD → Roon convert to PCM which I find interesting.

1 Like

In this case, also note that the algorithm used for the conversion makes difference. It is a very delicate process, even more than upsampling.

1 Like

So when I remembered my Sony Discman in highschool had a 1bit DAC I got to Googling about it and randomly landed on this old DIY audio thread.

This post by Thorsten Loesch (formerly AMR and iFi Audio fame but I think he’s left now?) is interesting:

What is a 1-bit DAC ? - Page 4 - diyAudio

Especially the part of Stereophile measurements showing noise at approx -50dB at 100kHz with some SACD players. He later explains that this noise may upset some amplifiers that create noise lower down the audible range.

But anyway given the age of that thread, it’s clear they are talking about 2.8MHz bitrate (SACD / DSD64).

This reminded me to go look at the Andreas Koch: Raising the Sample Rate of DSD - Is There a Sweet Spot? - Positive Feedback

And indeed for ~2.8MHz rate at 100kHz we see noise at approx -50dB. So that’s consistent with the old Stereophile measurements mentioned in that old post by Thorsten.

But for ~11.2MHz (DSD256 rate) that noise at 100kHz is around -125dB in that AK Figure 1. So much less an issue up to 100kHz compared with DSD64/SACD rate.

Maybe that noise is even lower with HQPlayer 1bit ~11.2MHz especially with ASDM7EC modulator @jussi_laako ?

This is much less of an issue than what happens with most PCM DACs, where the image frequencies are fully correlated with the signal. That noise from a good modulator is totally uncorrelated and would sound like a hiss, like tape hiss you get with old C-cassette (without Dolby noise reduction) or FM-radio. Unlike the correlated IMD-like distortion from PCM DACs.

How much noise you have in the DAC output depends on how the DSD to analog conversion is done (number of elements, analog post-filter, etc) and the used modulator. For example with Holo Spring, RME ADI-2 or iFi iDSD micro and ASDM5/ASDM7 (EC or non-EC) at DSD256, noise levels within 100 kHz are below analog noise floor.

1 Like

As I said before, DSD quantization noise cannot be uncorrelated. It’s simple math.

Yes it can, it can scrambled, similar to encryption. Care to open up the simple math?

While PCM images are directly correlated with source, repeating around every multiple of the sampling rate.

Just for fun, for example Chord Mojo has higher output quantization noise with PCM inputs than Holo Spring running at DSD256. In addition, Mojo’s modulator suffers from spurious tones and noise modulation…

As usual, it all depends on the modulator implementation. You can make a poor one, or a good one.

2 Likes

Since DSD samples are either +1 or -1, the power of the signal is always constant and equal to 1. When there’s no signal, all power is noise. When you add signal, noise power has to go down since the same amount of power is now shared between signal and noise. Thus, noise power is modulated by signal power. This has nothing to do with the modulator, it’s simply inherent to 1-bit coding and a result of the fact that it can’t be fully dithered (which is the only way to completely separate noise from power). PCM does not have this problem since you have a few bits to work with and full dithering does not saturate the quantizer.

Yes it can be fully dithered and yes it has to do with the modulator. For any SDM, the performance is completely dependent on the modulator algorithm. You cannot make statements based on just number of bits or sampling rate.

If PCM doesn’t have the problem, also DSD doesn’t have the problem, because DSD 0 dB equals to -6 dBFS. It is easiest to explain it this way; in PCM you can encode fully dithered signals that are below LSB level. Meaning that no other bits than your LSB ever changes. Then you can effectively throw away your unused MSBs.

If you are worried about DSD, also stay away from one 1-bit systems like class-D amplifiers, microphones on mobile phones, etc… Those are much simpler than advanced modulators.

4 Likes

First, let me clarify that what I call “full dither” is random dither with an amplitude of one delta (i.e. going between -delta and +delta), where “delta” is the quantization step.

If you keep only the LSB in PCM (which uses mid-thread quantization), you’re left with a signal with 3 levels (-1, 0, +1). That’s one more than DSD, and it can still represent zero (i.e. can have zero power).

Now let’s say you use a mid-riser quantizer like DSD. In that case, you can’t fully dither a non-zero signal without going over +delta (or under -delta) every now and then, which would be mapped to +1.5 delta level, i.e. second positive level (or to -1.5 delta, i.e. the second negative level). You’ll end up with four total levels, i.e. (-1.5, -0.5, +0.5, +1.5). Now that’s two more than DSD. You’d have to halve the amplitude of dither to stay with two levels.

And that works only if you don’t shape. If you add noise shaping on top, you have to go with an even lower dither amplitude since the addition of error will surely go over these values. You end up with a fraction of what can be called “full dither”.

Yeah, that’s fine. Although it not really applicable concept in SDM, or noise-shaped PCM. You don’t seem to understand how modulators can be dithered.

No, keep it unsigned, it doesn’t matter. Any dithered or noise-shaped signal will never have zero power signal. Even less so any analog source through a decent ADC.

Now you are making assumptions about the modulator…

That is actually fine, since DSD specification allows up to +3 dB levels.

Error amplitude doesn’t actually matter at all.

Just fun reminder, ESS DAC chips also use 1-bit stream from their HyperStream modulator. :wink:

1 Like

This is all fascinating stuff but…
I do have a suggestion for @Marian .

Please release your very own version of HQplayer, perfect it, sell it to many consumers, support it and continually fine tune and update it.

Then let’s see how level the field is.

Yes I realise this will get deleted by mods pretty quickly but …:thinking:

2 Likes

Seriously? Do we need credentials to contribute here?

1 Like

I’d like to know which part I don’t understand. To the best of my knowledge, dithering is an integral part of the quantizer, which is independent of any modulation done around it. Thus, one can speak of dithering in the context of a quantizer, not a modulator.

1 Like