The Ear - DSD under Fire

[mansr]

That's true, which I've pointed out before. I have no knowledge of the percentage of each in terms of unwanted data, and in my business it's not very relevant to me. What is relevant is the necessity of converting from PDM front ended A/D converters into PCM for much of anything more complex than pure editing. It's the conversions from one format to the other that damages the sound quality of either, and the requirement to do so in order to do post processing is dumb.

 

I suppose mathematics is dumb then.

[jabbr]

That's true, which I've pointed out before. I have no knowledge of the percentage of each in terms of unwanted data, and in my business it's not very relevant to me. What is relevant is the necessity of converting from PDM front ended A/D converters into PCM for much of anything more complex than pure editing. It's the conversions from one format to the other that damages the sound quality of either, and the requirement to do so in order to do post processing is dumb.

 

It only damages sound quality if the sampling rate and bit depth are not high enough.

[Hiro]

I suppose mathematics is dumb then.

 

There are smart ways to process DSD without PCM decimation and downsampling. And then there are production workflows that do not require any digital sweetening at all:

 

- acoustic recordings made direct to DSD

- acoustic recordings with on-site analog mixing & native DSD editing

- direct transfers from analog tapes

- all projects that use analog mixing consoles

[mansr]

Not following you along these lines. With PCM at an equivalent sample rate, you still encode an enormous amount of unwanted information.

 

There's a significant difference between redundant information and wrong information. Redundant information, as with PCM, can be easily removed using compression algorithms.

[Hiro]

That's not the whole story. It is correct that PCM samples carry a lot of redundant information. However, with PDM, you are necessarily encoding a vast amount of unwanted information, aka noise.

 

Isn't it the point of delta-sigma modulation to move the noise out of the audio band where it can get filtered out?

[mansr]

Isn't it the point of delta-sigma modulation to move the noise out of the audio band where it can get filtered out?

 

I thought you didn't like filters.

[Hiro]

I thought you didn't like filters.

 

I suspected that you weren't telling the whole story.

[mansr]

I suspected that you weren't telling the whole story.

 

Huh? You need a filter to get rid of the ultrasonic DSD noise. There's no way around it.

[Hiro]

Huh?

 

I mean when you were talking about the unwanted noise in delta sigma modulation.

[tailspn]

It only damages sound quality if the sampling rate and bit depth are not high enough.

 

No, that's not the real world, it's the ratios. To perform DXD processes on 64fs DSD content in DXD (352KHz PCM) it's an 8:1 fold down. With a Nyquist frequency of 176KHz, that's a pretty steep filter from 2.8MHz to 172KHz (4 octaves) in which to drop 120dB. It's easy to hear the results of that if you listen to instrument detail and spaciousness cues of acoustic music.

[mansr]

I mean when you were talking about unwanted noise in delta sigma modulation.

 

Even if it's outside the audible band, it's still taking up information space.

[Hiro]

Even if it's outside the audible band, it's still taking up information space.

 

When you shift the noise beyond 60kHz (in case of DSD128) that's quite an information space at your disposal.

 

Tom's point was that the vast amounts of data carried by PCM is redundant stuff, so simple bit counting doesn't make much sense when comparing PCM and PDM/SDM.

 

But the number of bits per unit time do not define the same total information in PDM and PCM, as applied to their use in audio. From each PCM sample to the next, all the constant information in addition to the change information is carried forward. That's conservatively at least 80% redundant data, signal rate of change dependent. Probable more like the high 90's percent.

[mansr]

When you shift the noise beyond 60kHz (in case of DSD128) that's quite an information space at your disposal.

 

Tom's point was that the vast amounts of data carried by PCM is redundant stuff, so simple bit counting doesn't make much sense when comparing PCM and PDM/SDM.

 

And DSD carries useless noise, so simple bit counting doesn't make much sense from that side either.

[Hiro]

And DSD carries useless noise, so simple bit counting doesn't make much sense from that side either.

 

We've already established that the noise is actually moved out of the way with delta sigma modulation.

[mansr]

We've already established that the noise is actually moved out of the way with delta sigma modulation.

 

And I keep telling you that it still uses some of the available information space provided the raw bit count. Since you appear immune to this knowledge, I won't repeat it again.

[Hiro]

And I keep telling you that it still uses some of the available information space provided the raw bit count. Since you appear immune to this knowledge, I won't repeat it again.

 

Yes, some, above 60kHz. And in the meantime Tom states that as much as 80% of PCM data is redundant.

[Tony Lauck]

That's not the whole story. It is correct that PCM samples carry a lot of redundant information. However, with PDM, you are necessarily encoding a vast amount of unwanted information, aka noise. The amount of useful information is comparable between the two encodings.

 

Unwanted information in multi-bit PCM is easily removed losslessly using adaptive prediction methods such as FLAC, which uses LPC prediction and sends for each sample the difference between the predicted and actual value (further compressed by an arithmetic code). For music this typically yields about 40% savings in bits. Similar levels of compression with 1 bit formats created by sigma-delta modulators are not possible because of the chaotic nature of the modulation, unless the sound quality is degraded by adding additional constraints on the bit stream, as in trellis coding.

[Tony Lauck]

Yes, some, above 60kHz. And in the meantime Tom states that as much as 80% of PCM data is redundant.

 

With DSD64 the noise is moved partially out of the way, but remaining noise still overlaps with musical information. If this noise is filtered out (to avoid problems with some equipment) then the required filtering is going to be no more transparent than the required filtering needed for 192 Khz PCM. Neither format can be said to be transparent, but of course it is possible to produce good quality recordings in both formats.

[Hiro]

Unwanted information in multi-bit PCM is easily removed losslessly using adaptive prediction methods such as FLAC, which uses LPC prediction and sends for each sample the difference between the predicted and actual value (further compressed by an arithmetic code). For music this typically yields about 40% savings in bits. Similar levels of compression with 1 bit formats created by sigma-delta modulators are not possible because of the chaotic nature of the modulation, unless the sound quality is degraded by adding additional constraints on the bit stream, as in trellis coding.

 

Apparently even larger amounts of redundant PCM data can be compressed using MQA, though it's not clear at the moment whether it isn't causing any degradation to the signal.

[mansr]

Apparently even larger amounts of redundant PCM data can be compressed using MQA, though it's not clear at the moment whether it isn't causing any degradation to the signal.

 

MQA is not lossless.

[Hiro]

With DSD64 the noise is moved partially out of the way, but remaining noise still overlaps with musical information. If this noise is filtered out (to avoid problems with some equipment) then the required filtering is going to be no more transparent than the required filtering needed for 192 Khz PCM. Neither format can be said to be transparent, but of course it is possible to produce good quality recordings in both formats.

 

When we talk about DSD today we should focus on DSD128 and DSD256 as they are supported by a growing number of DSD DACs, and record labels that offer DSD downloads are likewise moving to Double DSD and above. The latest Cat Stevens DSD download offered by Analogue Productions was done with Sony Pro DSD 128 A/D. Should be a beginning of a new Double DSD series. The new store Vision DSD is offering everything in 5.6MHz DSD. Opus3 records also to DSD128 to give a few examples.

[Tony Lauck]

When we talk about DSD today we should focus on DSD128 and DSD256 as they are supported by a growing number by DSD DACs.

 

Agreed. As with Redbook CD, Sony made the mistake of releasing a technology prematurely and as a result with too low a sampling rate.

[Tony Lauck]

I hadn't planned to respond, since it's off topic, but is your function versus mechanism statement meant to refer to Section 112 indefiniteness and means-plus-function language, a fraught area of the law on which the federal circuits are split, or something else?

 

My function vs. mechanism comment was based on my experience as a computer network architect with a computer science/mathematics background. I note that similar distinctions are in use in biology and ecology, among other fields. My experience as an expert/consultant in computer networking patent disputes came later. The basic problem is that most people do not handle abstractions willingly, and this even extends to many engineers.

 

Function: Draining the swamp.

Mechanism: keeping one's ass away from alligators.

[Hiro]

Agreed. As with Redbook CD, Sony made the mistake of releasing a technology prematurely and as a result with too low a sampling rate.

 

It would be great if they could provide support for higher DSD rates from the start, but it's typical for any technology to advance with time. Can anyone say that people who introduced USB 1.0 released the technology prematurely, because we then moved to USB 2.0, and more recently USB 3.0, with each offering faster transfer speeds?

[Tony Lauck]

 

Obtaining a sampled signal in a given representation is another matter. If the precision of the sample values is sufficient, simply rounding the real value to the closest representable value is good enough, the rounding error being a form of non-linear distortion. As the precision is decreased, this distortion increases. At 8-bit precision, simple rounding still produces recognisable audio although the distortion is also obvious. At 1-bit precision, a sine wave is turned into a square wave, and any finer details are mangled beyond recognition.

 

I got noisy but undistorted sound out of 96/8 PCM by using subtractive dither. There was no recognizable distortion on the music, but there was an obvious amount of benign white noise, about at the level of a typical 4 track prerecorded tape from the 1960's.