The sound quality of DSD?

[watchnerd]

MQA is a scam

FIFY

 

[Ron Party]

Could have sworn I've seen him post recently on this topic (like within a week or so), kept saying how wrong it was. My memory's hazy atm if the topic was actually DSD or not.

Ha ha, my memory isn't what it used to be, but it is not that bad! The more I think about it, he posted at AVS about this, and that was several years ago.

 

[ZolaIII]

I already knew that, but it was a good refresh of the topic, also thanks to you.

But it was pretty hard to find a more explicit explanation, but thanks to all. Just interpolate and reduce the wordlength of PCM and you get PDM, then apply noise shaping (using the error feedforward method you described) to move most of the quantization error energy out of the passband, and you get DSD. Interesting...

@pkane, I have a question: how do you increase word length in the conversion from PDM to multi-bit PCM, I know you have to decimate, but you do you recover the remaining "bits" out of an 1-bit signal? I don't find this anywhere, but I would like to know.

DSD 64 = PCM 176400 16 bit (perfect), PCM ≠ DSD (edit; this is about proper conversation from one to another companion).
PCM is transparent and editable, DSD is not.
Delta sigma modulation does add distortion (in predictable and controlled manner) which goes up with frequency in order to resolve this the filter which acts as a nail to tighten and polish hear able range is used. This filter creates an nus effect that the very small part of vibration from high ultrasonic range transfers back to the source (al the way back). This acts as an bolding & sharpening edges effect while all together distortion is kept under 1% in whole hear able range. Most of people tend to like this effect especially in lows even its smallest there.

 

[mansr]

As I presented evidence to back up my claims

You linked to a Wikipedia article. As it happens, that Wikipedia article is one of the most confusing jumbles I have ever read.

I have a 500-page book on the topic, Understanding Delta-Sigma Data Converters, that doesn't mention PDM or "pulse density" even once. Somehow, I trust three professors more than some random Wikipedia article and a few audio salesmen.

 

[mansr]

DSD 64 = PCM 176400 16 bit (perfect), PCM ≠ DSD.
PCM is transparent and editable, DSD is not.
Delta sigma modulation does add distortion (in predictable and controlled manner) which goes up with frequency in order to resolve this the filter which acts as a nail to tighten and polish hear able range is used. This filter creates an nus effect that the very small part of vibration from high ultrasonic range transfers back to the source (al the way back). This acts as an bolding & sharpening edges effect while all together distortion is kept under 1% in whole hear able range. Most of people tend to like this effect especially in lows even its smallest there.

Uh-huh.

 

[bennetng]

PCM is transparent and editable, DSD is not.

Interesting that no one show up again and ask for listening test results after witnessed a successful 44.1k vs 88.2k PCM ABX test. PCM-DSD_Converter supports DSD16 encoding, so it has identical bitrate as 16/44 PCM and here are some apple to apple comparison using a 5-tone 32-bit float 44.1kHz input signal. I converted the DSD16 file to PCM again with minimal high frequency droop, and therefore the "ultrasonic" (>5kHz) noise is high. Also included 16/44 PCM versions with flat and shaped dither for comparison.

 

[pkane]

@pkane, I have a question: how do you increase word length in the conversion from PDM to multi-bit PCM, I know you have to decimate, but you do you recover the remaining "bits" out of an 1-bit signal? I don't find this anywhere, but I would like to know.

Number of bits isn't the issue, SNR is. SNR determines the effective number of bits of resolution. In DSD/SDM encoding this is a function of the sampling rate and the order of the modulator. No matter how many bits you convert the signal into, the recorded SNR will remain the limiting factor.

 

[mansr]

Perhaps this is helpful: https://troll-audio.com/articles/pcm-and-dsd/

I have updated the article with some more graphs and explanations.

 

[KeithPhantom]

I have updated the article with some more graphs and explanations.

It made sense after I read Audio Precision's take on this subject. It isn't actual PDM as you said, the number of pulses (samples) has to be interpolated so when they are averaged it can represent the actual signal (or close to it), isn't that what is actually happening? Also, why not add an image that represents the delta modulation to help to visualize how the pulses are related to this type of modulation?

 

[mansr]

Also, why not add an image that represents the delta modulation to help to visualize how the pulses are related to this type of modulation?

Because delta modulation is something else.

 

[KeithPhantom]

Because delta modulation is something else.

Ummm, so DSD doesn't use DS?... Interesting, I'll see if I can find something related to that

 

[Jimbob54]

For me, that Tidal charges the same for regular lossless and MQA is a scam, they should charge less for just 44.1/16.

Leaving aside the merits or otherwise of MQA, they don't charge "the same" for lossless and MQA. They have a "Hifi" tier which is lossless (plus). The (plus) includes MQA. Whilst the % of their catalogue available in MQA is increasing, a lot of that is 44.1/24 or 48/24 MQA. Its not enough to justify opening up a new top tier.

You would be mad to pay extra for the "masters/mqa" tier . Its not broad enough.

Youre paying the same , yes, and I'm sure the MQA licensing costs is factored into that price but I dont see them splitting the top tier anytime soon.

 

[mansr]

Ummm, so DSD doesn't use DS?... Interesting, I'll see if I can find something related to that

Delta modulation is (sort of) half the delta-sigma loop. It is rarely, if ever, used any more, and discussing it wouldn't help the understanding of DSD.

 

[KeithPhantom]

Delta modulation is (sort of) half the delta-sigma loop. It is rarely, if ever, used any more, and discussing it wouldn't help the understanding of DSD.

Aren't most of the DACs oversampling multibit nowadays (they use like 5-6 bits, being pretty much oversampled PCM to make up for the apparent lack of bit depth)? I also think that multi-level noise shaping techniques are prevalent, but I would like to your confirmation before settling with this knowledge.

 

[mansr]

Aren't most of the DACs oversampling multibit nowadays (they use like 5-6 bits, being pretty much oversampled PCM to make up for the apparent lack of bit depth)? I also think that multi-level noise shaping techniques are prevalent, but I would like to your confirmation before settling with this knowledge.

Sigma-delta modulation can be used at any bit depth. If only gentle noise shaping is needed, simpler methods are sufficient, so SDM is not usually encountered at higher bit depths.

 

[KeithPhantom]

Sigma-delta modulation can be used at any bit depth.

I've read something about multi-level delta-sigma, which has multiple integrators by design, is this what we are talking about when SDM is used in multi-bit implementations?

 

[MRC01]

This might be a dumb question, but it will help me understand how DSD works, so here goes...

If DSD encodes 0 and 1 whenever the amplitude is below or above zero, how does it encode the amplitude of the signal? For example when we change the volume of a waveform, that doesn't change any of its zero crossing points. More generally, I can devise many different signals that all have the same zero crossing points. And if DSD works like that, it seems it would sample them all the same.

I suspect that DSD must work differently. Instead of simply being 0 or 1 when the signal is below or above zero, somehow, the ratio of 1s to 0s over a particular interval indicates the average amplitude over that interval. But this would mean that sometimes it encodes a 0 when the signal is above zero, and a 1 when it's below zero. Put differently, the probability of encoding a 1 or 0 would be proportional to the amplitude of the signal being encoded at that point. I don't know, just trying to make sense of it.

 

[mansr]

If DSD encodes 0 and 1 whenever the amplitude is below or above zero, how does it encode the amplitude of the signal? For example when we change the volume of a waveform, that doesn't change any of its zero crossing points. More generally, I can devise many different signals that all have the same zero crossing points. And if DSD works like that, it seems it would sample them all the same.

That's what you'd get without noise shaping, which is what sigma-delta modulation is all about.

I suspect that DSD must work differently. Instead of simply being 0 or 1 when the signal is below or above zero, somehow, the ratio of 1s to 0s over a particular interval indicates the average amplitude over that interval. But this would mean that sometimes it encodes a 0 when the signal is above zero, and a 1 when it's below zero. Put differently, the probability of encoding a 1 or 0 would be proportional to the amplitude of the signal being encoded at that point. I don't know, just trying to make sense of it.

That's pretty much what the result looks like. Here's a picture:

 

[tmtomh]

That is pretty much entirely wrong, if that.

Perhaps this is helpful: https://troll-audio.com/articles/pcm-and-dsd/

Extremely helpful - thanks!

After reading this, I'm left to ponder what the point of DSD is.

 

[KeithPhantom]

After reading this, I'm left to ponder what the point of DSD is.

Sell you the same music, since regular PCM can do everything DSD can.