The Multibit DSD debate

[Sam Lord]

Thanks tailspn. I know the system works, my best complex music recordings *by far* are SACDs (transfers from analog or hi-res digital), but what if there's a data dropout? The ground reference is lost--it could be estimated, but that's risky. There is metadata in the DSD stream for error correction, so I thought there might be a flag or something when zero-crossing occurs. And the Scarlet book spec doesn't require a series capacitor in the sending or receiving analog portion of the circuit AFAIK, so...how do they find the reference? Maybe the software determines when significant DC is found, then works back towards a AC signal, quite imperfectly.

[Miska]

And the Scarlet book spec doesn't require a series capacitor in the sending or receiving analog portion of the circuit AFAIK, so...how do they find the reference? Maybe the software determines when significant DC is found, then works back towards a AC signal, quite imperfectly.

 

Since input is ground referenced, and the bitstream has "virtual ground" at value 0.5, there's no problem. Generated bitstream will find the correct position after some tens of samples regardless where you cut it.

 

So if you make a simple switch output, create two ground-referenced voltage rails, -5V and +5V and then when driving a push-pull circuit '1' turns on the +5V rail switch and '0' turns on the -5V rail switch.

 

Overall distribution of 0's and 1's should be the same.

 

You could roughly think it as equivalent of using XLR with pin 1 disconnected. Especially if you feed it through a transformer (like is done in AES/EBU), you can re-reference the ground at secondary coil side.

[tailspn]

but what if there's a data dropout? The ground reference is lost--it could be estimated, but that's risky.

 

Yes, that why I said the effect is like that of a coupling capacitor in an analog circuit. There is no absolute ground reference in a PDM (DSD) data stream. As Jussi points out, the modulated signal (pulse density) that the PDM bit stream represents has a "virtual ground", which is the mid analog amplitude point of an average of a number of samples when integrated back into a pure analog signal.

 

The system operates on the basis that the analog signal is approximately symmetrically balanced about its mid level. That is certainly true on the long term (several cycles of the lowest frequency of interest), but is not true on the short term. The DC offset in a reconstructed (integrated) PDM bit stream is always wondering around, which is why there's a specification of -50dB for SACD production, and why you get a click when randomly slicing a file.

[Miska]

This of course also applies to delta-sigma ADCs with PCM output too, that's why practically all such ADC chips have DC-block high-pass digital filter. And real world implementations usually have either coupling capacitors in input, or DC servos (the way I like to do).

 

Just as an example two very usual pro-audio ADCs (notice "HPF" and "HP Filter" blocks)...

 

AK5394A:

 

CS5381:

[SoNic67]

I'm only speaking for myself here. It seems like you've declared victory and Berkeley Audio Design has conceded based on your statement. I think this is very misleading on your part.

That's why I stopped posting here. He and his 'followers' will never agree that they are wrong, even in the face of great arguments.

To say the he knows better then the guys that invented HDCD (Pacific Microsonics) is just ridiculous.

Sadly, I know that he has a vested commercial interest in all this and won't admit any other points of view.

[Miska]

That's why I stopped posting here. He and his 'followers' will never agree that they are wrong, even in the face of great arguments.

 

The technical data you have posted is quite scarce. Yeah, lot of arguing, but not much of technical data to back up the arguments.

 

To say the he knows better then the guys that invented HDCD (Pacific Microsonics) is just ridiculous.

 

That wasn't BAD wasn't it? And would a simple LSB coding + filter switching system from 18 years ago gain some special authority on something?

 

I never take anybody as authority as granted, I always make my own valuation. I don't care if it's President of Finland or not, just yet another human being.

 

(Yes, I don't respect authorities)

 

Sadly, I know that he has a vested commercial interest in all this and won't admit any other points of view.

 

Oh yeah and what is that? Show me the money!

[Hiro]

To say the he knows better then the guys that invented HDCD (Pacific Microsonics) is just ridiculous.

 

Steve Hoffman dumped HDCD as he found the encoding to have negative effect on sound quality. Anyway, to suggest that Miska or Michal Jurewicz aren't right just because they didn't invent hdcd is preposterous.

[Teresa]

Steve Hoffman dumped HDCD as he found the encoding to have negative effect on sound quality. Anyway, to suggest that Miska or Michal Jurewicz aren't right just because they didn't invent hdcd is preposterous.

 

I found an old post of mine on sa-cd.net from 2009

 

HDCDs are 20 Bit 44.1kHz but to gain the 4 additional bits the HDCD has to be properly decoded. According to Gene Pope of "Pope Music" undecoded HDCDs are only 15 Bit as the extra information is perpetually coded in the LSB (least significant bit).

 

"The extraordinary fidelity of the HDCD process is achieved by identifying and correcting previously misunderstood [or unknown] sources of distortion in digital audio reproduction. These include both additive artifacts of the analog-to-digital and digital-to-analog conversion processes, and subtractive distortions resulting from insufficient data present in the 44.1 kHz, 16-bit PCM sampling standard of the compact disc format. The HDCD process effectively cancels the additive distortions and simultaneously provides additional data to reduce the subtractive distortions."

 

Spalinger Reference hdcd

 

Everyone knows how much I hate CDs, well HDCDs can be actually quite good especially the ones from Reference Recordings. The difference between CD and HDCD in sonics and listenability is huge, however I no longer find the resolution fine enough to be acceptable since the advent of SACD and DVD-Audio. I have long since sold all my HDCDs as well as my Adcom GDA-700 HDCD DAC.

 

The SACD and the 24 Bit 88.2 and 96kHz downloads of Reference Recordings are vastly superior to the previous HDCD versions. However before high resolution digital I listened exclusively to LPs and HDCDs.

 

Since according to Gene Pope undecoded HDCDs are only 15 Bit could that be the reason Steve Hoffman dumped HDCD?

[tailspn]

Hi Teresa, long time....

[The Computer Audiophile]

Steve Hoffman dumped HDCD as he found the encoding to have negative effect on sound quality.

To be 100% honest, who cares.

[SoNic67]

The technical data you have posted is quite scarce. Yeah, lot of arguing, but not much of technical data to back up the arguments.

Sure. Except the fact that the whole DAC industry run away from 1 bit D-S because of it's obvious issues. That's why we have multilevel D-S today. Only you know better than them that 1 bit D-S is the future.

Like I said, I cannot imagine a more inefficient way to store the audio digitally. All that noise-shaping that NEEDS to be carried around! And then pushed into the first stage of an OpAmp filter and hope for the best!

[elcorso]

Steve Hoffman dumped HDCD as he found the encoding to have negative effect on sound quality. Anyway, to suggest that Miska or Michal Jurewicz aren't right just because they didn't invent hdcd is preposterous.

 

There is a lot of people that never liked HDCD, like me.

 

Bought then by Microsoft only to decrease his already down market share value...?

 

Roch

[Miska]

Sure. Except the fact that the whole DAC industry run away from 1 bit D-S because of it's obvious issues. That's why we have multilevel D-S today.

 

It doesn't take anything away from DSD. I am using discrete 33-level D-S converter to convert DSD to analog and it's 100% bit-perfect! (I designed it, so I know)

 

5-level D-S used by for example TI is able to have whopping 9.5 dB lower noise than DSD!

 

Now I'm just waiting for downloads I can buy in multi-bit D-S. Perfectly fine for me. I can also upsample to multi-level D-S so I can hardly wait to see DACs on the market that can take in native multi-bit D-S input at 24.576 MHz or higher rates. I'm ready!

 

Like I said, I cannot imagine a more inefficient way to store the audio digitally. All that noise-shaping that NEEDS to be carried around!

 

It has 1.4 MHz Nyquist-bandwidth at 2.8 Mbps bitrate. I would say it's pretty efficient.

 

And then pushed into the first stage of an OpAmp filter and hope for the best!

 

If you are doing it that way, you are doing it wrong.

[Hiro]

There is a lot of people that never liked HDCD, like me.

 

 

I know, many at SH forum welcomed his decision.

[Hiro]

Since according to Gene Pope undecoded HDCDs are only 15 Bit could that be the reason Steve Hoffman dumped HDCD?

 

He had the proper equipment but found an A/D transfer to sound more transparent without the additional HDCD processing in the chain.

[SoNic67]

Can you support this allegation?

What burried the HDCD was the advent of DVD-V and DVD-A that allowed 24 bit uncompressed LPCM.

Truth is that even today, there are artists using the HDCD ADC because they hear that none of the present day ADC can compare with the sound quality (they don't even use the HDCD dynamic range extended flags, they do it just for the quality of the digital filters).

It has 1.4 MHz Nyquist-bandwidth at 2.8 Mbps bitrate. I would say it's pretty efficient.

It has a REAL bandwidth of 30-40kHz. Rest is noise. It's pretty inefficient, especially when you dump all that noise in the next stage.

[The Computer Audiophile]

Can you support this allegation?

What burried the HDCD was the advent of DVD-V and DVD-A that allowed 24 bit uncompressed LPCM.

Truth is that even today, there are artists using the HDCD ADC because they hear that none of the present day ADC can compare with the sound quality (they don't even use the HDCD dynamic range extended flags, they do it just for the quality of the digital filters).

 

It has a REAL bandwidth of 30-40kHz. Rest is noise. It's pretty inefficient, especially when you dump all that noise in the next stage.

DVD-A didn't bury anything accept itself. DVD-Video has nothing to do with HDCD. People still use it today because they like the sound of the Pacific Microsonics Model One and Model Two. HDCD was never a market force. The two things that shrunk what little market share it had are 1. Microsoft purchasing Pacific Microsonics and killing off the main thrust of the technology, and 2. The discontinuation of the Pacific Microsonics ADC/DAC units. If you want to make an HDCD now you have to have one of the remaining Model Ones or Twos. The converters are hard to find and very expensive on the used market. Studios want less expensive gear, for the most part, and they don't want gear that may break down without an easy fix. It's possible to get them serviced though.

[Savage]

DVD-A didn't bury anything accept itself. DVD-Video has nothing to do with HDCD. People still use it today because they like the sound of the Pacific Microsonics Model One and Model Two. HDCD was never a market force. The two things that shrunk what little market share it had are 1. Microsoft purchasing Pacific Microsonics and killing off the main thrust of the technology, and 2. The discontinuation of the Pacific Microsonics ADC/DAC units. If you want to make an HDCD now you have to have one of the remaining Model Ones or Twos. The converters are hard to find and very expensive on the used market. Studios want less expensive gear, for the most part, and they don't want gear that may break down without an easy fix. It's possible to get them serviced though.

 

I'll flirt with OT, why did Microsoft buy HDCD and why did they bury it. Or, why did Berkeley sell it?

[dallasjustice]

I am using discrete 33-level D-S converter to convert DSD to analog and it's 100% bit-perfect!

 

It's not bit perfect when the sound hits your ears. The post electrical air filter always changes the bits and you can't compensate for that without converting DSD to another format. Other than the very cumbersome HQPLAYER, there's no other player that can do any DSP without converting to PCM. DSD sounds good in theory but in the real world it's a big diversion from high quality digital playback.

[The Computer Audiophile]

I'll flirt with OT, why did Microsoft buy HDCD and why did they bury it. Or, why did Berkeley sell it?

It's my understanding Microsoft had someone on staff who really understood audio and saw HDCD as a good product for Microsoft's portfolio. That person is no longer at Microsoft and nobody there has cared about HDCD for a number of years. It sounds crazy to many of us, but when a company gets that huge and has so many acquisitions, stuff like HDCD can just disappear.

 

By the way, Pacific Microsonics sold HDCD to Microsoft not Berkeley Audio Design.

[Hiro]

Can you support this allegation?

What burried the HDCD was the advent of DVD-V and DVD-A that allowed 24 bit uncompressed LPCM.

 

Turns out Stephen Marsh wasn't happy with HDCD filtering/processing either...

 

I've decided not to run my masterings through an HDCD converter.. | Steve Hoffman Music Forums

[The Computer Audiophile]

Turns out Stephen Marsh wasn't happy with HDCD filtering/processing either...

 

I've decided not to run my masterings through an HDCD converter.. | Steve Hoffman Music Forums

 

Turns out Grammy winning engineer Keith O Johnson is very happy with HDCD and the PM2 :~)

[Hiro]

Now I'm just waiting for downloads I can buy in multi-bit D-S. Perfectly fine for me. I can also upsample to multi-level D-S so I can hardly wait to see DACs on the market that can take in native multi-bit D-S input at 24.576 MHz or higher rates. I'm ready!

 

I'm waiting for digital amplifiers with switching speeds of 2.8MHz/5.6MHz/11.2MHz....

[Sam Lord]

Since input is ground referenced, and the bitstream has "virtual ground" at value 0.5, there's no problem.... equivalent of using XLR with pin 1 disconnected.... you can re-reference the ground...

 

Miska, Once again you have very kindly explained some basic electronics to the benefit of everyone, in this case some basics of binary digital signal transmission. But I didn't phrase my question well, and with that realization and your answer, it led me to the proper question which I can find the answer to.

 

I'm very familiar with what you described, I have lots of basic digital signal experience. What I intended to ask about was a problem that is not addressed except in iLink and SDIF and DoP specifications. I didn't intend to ask about the physical integrity of the binary code when an error of dropout occurred. I was asking about the absolute level of the converted analog signal when the DSD bitstream is interrupted.

 

Each discrete level change in DSD is completely dependent on *all* of the preceding bits of a track, unless there is metadata to indicate zero crossing or silence. And there is a silence signal, only partly defined in Sony literature as an 8-bit segment with an even number of ones and zeroes. But no ground reference is mentioned in any available DSD literature I could find. The problem: On bit # 295,365,762 of a DSD track on a given channel, the reconstructed output level, representing a specified voltage, depends fully on the difference in the number of ones and zeroes that preceded that bit. If your SACD has a big scratch, you have lost your correct level. I don't know the minimum corner frequency for DSD, maybe it's 5Hz, so that does mandate the elimination of DC offset by the converter system. But does the converter have to hunt about for correct level for a while, or is a zero crossing point marker mandated?

 

OK, then I realized that this is not really a DSD problem, because by Sony edict DSD is transmitted *internally*, i.e. *within a chassis* and has vanishingly low transmission error rates. It is however something which *probably* is addressed in external iLink and DoP transmission of DSD data. Because there *will* be dropouts which CRC codes cannot reconstruct. DSD data frames occur 75 times per second, so they will not necessarily contain code for a zero crossing. So I need to look at the DoP and iLink specifications, I'll bet there are absolute level markers somewhere in those datastreams.

 

Yes, that why I said the effect is like that of a coupling capacitor in an analog circuit. There is no absolute ground reference in a PDM (DSD) data stream... ...The DC offset in a reconstructed (integrated) PDM bit stream is always wondering around, which is why there's a specification of -50dB for SACD production, and why you get a click when randomly slicing a file.

 

Agreed there is always a requirement in any (intended) AC-coupled system to have analog and/or digital circuitry to reset DC to reasonable level. -50dB ref 2V is what, about 3mV? Sensible. But I can't believe that when the stream is transmitted for several feet or over the internet there isn't metadata to denote zero crossings or absolute level every second or so. Otherwise the converter will hunt for the correct level all the way to the end of a track in some cases.

 

This of course also applies to delta-sigma ADCs with PCM output too, that's why practically all such ADC chips have DC-block high-pass digital filter. And real world implementations usually have either coupling capacitors in input, or DC servos (the way I like to do).

Just as an example two very usual pro-audio ADCs (notice "HPF" and "HP Filter" blocks)...

 

Thanks for the diagrams Miska. Neither of those analog or digital filters will accomplish the restoration of exact intended signal level without a fairly long computation tail. So I'm still convinced for reasons above that metadata that somehow provides accurate level at regular intervals is part of the DoP and iLink specifications. And I expect this need will exist for any multibit SDM transmission of several seconds. But hey I can be surprised.

 

Thank you Miska and tailspn for your responses, and I'm sorry for taking the thread off topic (a less-than endearing talent of mine).

[Superdad]

Thank you Miska and tailspn for your responses, and I'm sorry for taking the thread off topic (a less-than endearing talent of mine).

 

Please don't apologize Sam. This discussion is fascinating wherever it leads. I think John (Swenson) recently dealt with the exact issue you raise during a DSD project. I am going to forward your post to him since he may have some very good answers.

--Alex C.