To DSD or not to DSD?

[k6davis]

Well, what's inside a DSD-only Lampizator then?

 

Exactly. Because I can attest that it sounds fantastic.

[k6davis]

Returning to the fundamental point that should not be forgotten, the quality of the original recording in terms of mike setup, etc. is far more important than the format, be it PCM or DSD.

 

This point is made often, and it should be, because it's certainly true.

 

What I find more interesting though, is that I greatly enjoy the sound of all of my music (PCM of all bitrates.. and even DSD64.. and even the few lowly mp3 files that I haven't found in a better format) upsampled to DSD128. When Lampizator offers DSD256, which will be soon, I'll upgrade and try that too.

 

In that scenario, differences in the quality of the original recording, mastering, type of music, system and room are all negated and DSD is the best sound I've heard - in my humble opinion, of course.

 

I respect those who feel that such DSP manipulation is heresy, but I love the way it sounds. So much so, that I bought a Lampizator DAC without PCM decoding - which saved me a bit of cash to boot.

 

It's difficult at best to have meaningful discussions on this and other audiophile topics when we all have completely different systems - and rooms, of course. I think that differences among system and space are just as significant as differences in individual preference, maybe even more so.

 

We have these in-depth conversations, but "DSD" definitely doesn't sound the same in my house as it does in yours. I auditioned about 4 DACs in-house before I chose one and they all sounded quite different, not to mention the even greater variances in speakers and all of the rest of it.

 

It's a shame that there's no way for us to hear each others systems and *then* discuss. Imagine how enlightening that would be!

[Superdad]

Well, what's inside a DSD-only Lampizator then?

 

Well, nobody knows what kind of filter it uses since it is sealed in a black box module. Plus lots of tubes, caps, wood, and maybe even an output transformer: LampizatOr DSD DAC

 

I did not say it could not be done--John's prototypes were magical in my system. But aside from eschewing a standard DAC chip, I would hardly classify Mr. Fikus' design as minimalist. Indeed there is much to be heard in SD/PDM recordings, be they 1-bit DSD or multi-level SDM conversions within DACs. I trust that may designers will continue to explore new topologies to reveal much more from good recordings. I doubt I'll ever own another DAC with an off-the-shelf DAC chip--at least not for my reference system.

 

Regards,

--Alex C.

[Superdad]

It's a shame that there's no way for us to hear each others systems and *then* discuss. Imagine how enlightening that would be!

 

+1

[Maldur]

Right, but with DSD you just need to pass it through an integrator (to produce an analog waveform) and then a low-pass filter (to remove high-frequency information)... both which are simple RC circuits, no?

 

About integrator... I don't know, when I experimenting with Amanero interface, I needed only RC circuits for RF noise, with cooperation of analog amplifier's input filter results are listenable (actually, very listenable), so called magic is there. CS4382 in DSD mode with Direct Out is less noisier, but I feel that magic is weak :).

 

But yes, as Superdad pointed out - some weird modulated low level noises can occur, not all tracks but some. It depends about overall recording level and I think, of ADC modulator properties (how it uses noise shaping during recording).

 

So, shift-registers are better starting point. I have Miska's DSC1 schematics on the table right now. And similar works from Japan (from Bunpei's friend).

[spdif-usb]

Returning to the fundamental point that should not be forgotten, the quality of the original recording in terms of mike setup, etc. is far more important than the format, be it PCM or DSD,

 

I enjoy music in both formats but, were I forced to choose, I would go with DSD. My Luxman DAC sounds absolutely glorious with good DSD material.

According to Barry Diament, the quality of the original recording in terms of mike setup, etc. can fully be had from PCM done properly at 4x rates (24/176.4 or 24/192), i.e. he has testified that it is audibly indistinguishable from the live feed in his studio. Also according to him, IIRC the same cannot be said about DSD, and, based on my own, personal experience, I have to perfectly agree with him.

 

That being said, one of the most plausible explanations as to why this is so IMO is the long established fact 1-bit quantizers, when properly dithered, are in constant overload. The resulting problems are distortion, noise modulation, and idle tones.

 

While it is true that choosing a higher order Sigma-Delta modulator shifts more noise out of the audible band, it is also true that there exists an optimum order when designing such a modulator, due to practical constraints. However, oversampling not only shifts even more noise out of the audible band, but also shifts it farther above the audible band, where it can be more effectively filtered.

 

On top of that, decimation increases accuracy by means of averaging, i.e. decimation does not in any way imply that redundant sampled values are merely being discarded. So in modern audio ADCs outputting PCM, there is usually a more complex combination of the digital filter and several other optimizations. The idea that properly done DSD is more accurate or more close to the original analog input signal than properly done PCM is a myth.

[Superdad]

But yes, as Superdad pointed out - some weird modulated low level noises can occur, not all tracks but some. It depends about overall recording level and I think, of ADC modulator properties (how it uses noise shaping during recording).

 

That's exactly it! On a lot of tracks, both native DSD and if I recall even some converted from PCM, during very low level passages (but not during silence), this strange noise--a bit like the sound of frying bacon--is heard. But it is not there during all quiet passages-- a single instrument or vocal is enough to stop it.

John Swenson explained to me what this "pattern noise" is and theories on what it occurs, but I have the memory of a goldfish for heavy tech stuff.

 

 

By the way, I was just reading on the Lampizator DSD DAC page, and to quote the designer directly: "THE LAMPIZATOR DSD DAC HAS USB PORT BUILT IN, SOLID STATE DIGITAL FILTER, PASSIVE DISCRETE ANALOG FILTER AND ACTIVE DISCRETE TUBE FILTER."

 

So for all we know, he too is using a shift-register followed by passive and tube low pass filters.

[wisnon]

Well, nobody knows what kind of filter it uses since it is sealed in a black box module. Plus lots of tubes, caps, wood, and maybe even an output transformer: LampizatOr DSD DAC

 

I did not say it could not be done--John's prototypes were magical in my system. But aside from eschewing a standard DAC chip, I would hardly classify Mr. Fikus' design as minimalist. Indeed there is much to be heard in SD/PDM recordings, be they 1-bit DSD or multi-level SDM conversions within DACs. I trust that may designers will continue to explore new topologies to reveal much more from good recordings. I doubt I'll ever own another DAC with an off-the-shelf DAC chip--at least not for my reference system.

 

Regards,

--Alex C.

 

Lampi does NOT use an output transformer..That I know for sure. The rest I cant say for certain. And yes, its as minimalist as can be and more minimalist than the rest as far as I can see.

[wisnon]

 

 

By the way, I was just reading on the Lampizator DSD DAC page, and to quote the designer directly: "THE LAMPIZATOR DSD DAC HAS USB PORT BUILT IN, SOLID STATE DIGITAL FILTER, PASSIVE DISCRETE ANALOG FILTER AND ACTIVE DISCRETE TUBE FILTER."

 

So for all we know, he too is using a shift-register followed by passive and tube low pass filters.

 

Dunno, bit it sure does sound good. Ask Bruce from Puget Sound!

[matthias]

And similar works from Japan (from Bunpei's friend).

 

Maldur,

can you post a link?

 

Thanks

Matt

[Maldur]

That schematic was in DIYAudio forum, if I remember correctly - USB to I2S 384Khz - DSD Converter - diyAudio

[Allan F]

According to Barry Diament, the quality of the original recording in terms of mike setup, etc. can fully be had from PCM done properly at 4x rates (24/176.4 or 24/192), i.e. he has testified that it is audibly indistinguishable from the live feed in his studio.

 

I didn't know that he offered that opinion to a court or some other official body. You will get different opinions from DSD advocates such as Cookie Marenco of Blue Coast Records and Jan-Eric Persson of Opus 3. As Andreas Koch, principal of Playback Designs and one of the developers of DSD frequently points out at seminars, the choice of PCM or DSD is primarily an artistic decision made by the producer, and that is why both formats should be supported. His philosophy is a breath of fresh air compared to the DSD naysayers in the audio industry.

[Miska]

That being said, one of the most plausible explanations as to why this is so IMO is the long established fact 1-bit quantizers, when properly dithered, are in constant overload. The resulting problems are distortion, noise modulation, and idle tones.

 

That happens when the modulator is NOT properly dithered, and it doesn't matter if it's 1-bit or any 1+N -bit. Especially with low-order modulators.

 

I have been very careful to design my modulators and the dithering scheme and I have been analyzing and measuring those a lot, as well has bunch of hardware multi-bit modulators inside DAC chips.

 

Most of the time DAC chips perform badly when converting from PCM input because 1) they have inadequate oversampling filters, 2) they use low-order modulators.

 

On top of that, decimation increases accuracy by means of averaging, i.e. decimation does not in any way imply that redundant sampled values are merely being discarded. So in modern audio ADCs outputting PCM, there is usually a more complex combination of the digital filter and several other optimizations.

 

Too bad most of those ADC decimation filters are also bad, offering only something like 100 - 120 dB stop-band attenuation. I prefer to use >200 dB...

 

Decimation is just a useless step because eventually you'll need to oversample and modulate again for the playback. Plus you get time domain effects of the decimation filter.

 

The idea that properly done DSD is more accurate or more close to the original analog input signal than properly done PCM is a myth.

 

PCM won't be "properly" done as long as it happens inside a tiny $10 chip. Very likely not the DSD either...

[Maldur]

That happens when the modulator is NOT properly dithered, and it doesn't matter if it's 1-bit or any 1+N -bit. Especially with low-order modulators.

 

Hi, Miska. As I wrote earlier:

But yes, as Superdad pointed out - some weird modulated low level noises can occur, not all tracks but some. It depends about overall recording level and I think, of ADC modulator properties (how it uses noise shaping during recording).

 

So this low level noise is originated from (ADC) modulator, I am right?

[Superdad]

Maldur,

can you post a link?

 

Thanks

Matt

 

Pretty sure he meant this one:

http://fpga.cool.coocan.jp/electrart/FIRTST.pdf from Mr. Motoi Tariki of ElectrArt (Electrart SHOP).

 

I think Miska's fronted by a good USB board will sound better. Mr. Tariki uses the transformer-based ADUM isolators which may cause more trouble than they cure in that application.

And both his and Miska's demonstration design would likely benefit greatly from a nicer output stage.

[Superdad]

That happens when the modulator is NOT properly dithered, and it doesn't matter if it's 1-bit or any 1+N -bit. Especially with low-order modulators.

 

I have been very careful to design my modulators and the dithering scheme and I have been analyzing and measuring those a lot, as well has bunch of hardware multi-bit modulators inside DAC chips.

 

Most of the time DAC chips perform badly when converting from PCM input because 1) they have inadequate oversampling filters, 2) they use low-order modulators.

 

Too bad most of those ADC decimation filters are also bad, offering only something like 100 - 120 dB stop-band attenuation. I prefer to use >200 dB...

 

Decimation is just a useless step because eventually you'll need to oversample and modulate again for the playback. Plus you get time domain effects of the decimation filter.

 

PCM won't be "properly" done as long as it happens inside a tiny $10 chip. Very likely not the DSD either...

 

 

Wow, that is a lot of wise truth packed into one post! Too bad the CA forum doesn't work like Quora or others where "up-votes" make the best posts rise to the top of the thread. Thanks Jussi.

[Miska]

So this low level noise is originated from (ADC) modulator, I am right?

 

Maybe, directly or indirectly... But it would need further analysis of the data and the DAC to say what is the root cause behind the symptom. Based on what Superdad said, could be limit cycles of the modulator and more audible due to problems in the converter design.

 

There are lot of converters that don't reach the required DSD noise filtering specs and this may cause the behavior to be clearly audible.

[One and a half]

I have no idea. If someone wants to send me one, with permission to open it up, I'd be glad to share what I found!

 

John S.

 

Some photos in Google images search , "lampizator interior photos".

[Maldur]

Pretty sure he meant this one:

http://fpga.cool.coocan.jp/electrart/FIRTST.pdf from Mr. Motoi Tariki of ElectrArt (Electrart SHOP).

 

Yes, exactly. I think I don't follow mr. Tariki's schematic fully, and that isolator can be some faster CMOS only chip (if needed). For USB interface I use definitely an Amanero board as Miska did, I have two of them on hand.

[spdif-usb]

That happens when the modulator is NOT properly dithered, and it doesn't matter if it's 1-bit or any 1+N -bit. Especially with low-order modulators.

I agree. However, if an ADC uses a Sigma-Delta modulator that is NOT 1-bit, IIRC that means its output cannot directly be stored in the form of DSD (i.e. due to the fact the DSD format is 1-bit). So, before it can be stored as DSD (please correct me if I am wrong), it needs to be re-quantized to 1-bit, and, to be able to do that, an 1-bit quantizer must be part of the design. This is a problem because, like I already said, an 1-bit quantizer cannot properly be dithered. If, on the other hand, the ADC uses a Sigma-Delta modulator that IS 1-bit, that means it also uses an 1-bit quantizer because an 1-bit Sigma-Delta modulator IS an 1-bit quantizer. So (again, please correct me if I am wrong) the problem persists both ways.

I have been very careful to design my modulators and the dithering scheme and I have been analyzing and measuring those a lot, as well has bunch of hardware multi-bit modulators inside DAC chips.

 

Most of the time DAC chips perform badly when converting from PCM input because 1) they have inadequate oversampling filters, 2) they use low-order modulators.

 

 

 

Too bad most of those ADC decimation filters are also bad, offering only something like 100 - 120 dB stop-band attenuation. I prefer to use >200 dB...

I don't doubt that. IIRC, Barry Diament also wrote that the number of ADCs and DACs that can perform well enough to be audibly indistuinguishable from the live mic feed is extremely small.

Decimation is just a useless step because eventually you'll need to oversample and modulate again for the playback. Plus you get time domain effects of the decimation filter.

IMO it's not useless because properly dithered means the amount of TPDF dither needs to be ±1 LSB, and that's not possible without increasing the word length. So there is still no free lunch I guess. Further, the time domain effects of the filter can, in human audible terms, be reduced by designing a better filter. With DSD, there also needs to be a filter to eliminate the enormous amount of quantization noise that results from the 1-bit quantizer. Plus oversampling shifts the quantization noise farther above the audible band, where it can be more easily filtered, more easily here meaning the filter can have less of the unwanted effects that you mention.

PCM won't be "properly" done as long as it happens inside a tiny $10 chip. Very likely not the DSD either...

Of course.

[Miska]

However, if an ADC uses a Sigma-Delta modulator that is NOT 1-bit, IIRC that means its output cannot directly be stored in the form of DSD (i.e. due to the fact the DSD format is 1-bit). So, before it can be stored as DSD (please correct me if I am wrong), it needs to be re-quantized to 1-bit, and, to be able to do that, an 1-bit quantizer must be part of the design. This is a problem because, like I already said, an 1-bit quantizer cannot properly be dithered. If, on the other hand, the ADC uses a Sigma-Delta modulator that IS 1-bit, that means it also uses an 1-bit quantizer because an 1-bit Sigma-Delta modulator IS an 1-bit quantizer. So (again, please correct me if I am wrong) the problem persists both ways.

 

Yes, so usually DSD recorders use a true 1-bit DSD ADC like PCM4202. (or a discrete design like Grimm AD1, and probably EMM Labs)

 

For multi-bit SDM it is still better to keep it as-is and not decimate at all.

 

IMO it's not useless because properly dithered means the amount of TPDF dither needs to be ±1 LSB, and that's not possible without increasing the word length.

 

Now you are talking about PCM... SDM is dithered in completely different way. Dithering is important when you reduce word-length. When you are decimating SDM to PCM you are doing word-length expansion.

 

So there is still no free lunch I guess. Further, the time domain effects of the filter can, in human audible terms, be reduced by designing a better filter.

 

Yes, you'll need to design two good filters, instead of one. There's no way to overcome all the limitations, you can get asymptotically closer, but never reach the optimum. And one of those is out of listener's control. And even worse, possibly located inside ADC chip where you cannot replace it afterwards with a better one.

 

Similar to digital cameras, you should be capturing the raw converter output when recording. Similar to using DNG with digital camera instead of JPG. Taking PCM out of modern ADC is analogous to storing to a JPG with DSLR.

 

With DSD, there also needs to be a filter to eliminate the enormous amount of quantization noise that results from the 1-bit quantizer. Plus oversampling shifts the quantization noise farther above the audible band, where it can be more easily filtered, more easily here meaning the filter can have less of the unwanted effects that you mention.

 

You need to do that with multi-bit SDMs too. DSD is of course no different in terms of oversampling and runs at similar rates compared to multi-bit SDMs (most run at same as DSD128). With PCM inputs to common DAC chips you in addition get side-effects of the technical limitations and thus get much stronger correlated image freqencies around multiples of 352.8/384 kHz. This for example results in IMD products at 2 kHz for 1 kHz input signal, using properly done DSD as input removes these...

 

When you have full chain in DSD, you need only relatively gentle filter to remove the output noise. Not something like the brickwall filter needed to decimate SDM to PCM. For DSD, you are fine with 1st order anti-alias filter for the ADC.

[Boris75]

According to Barry Diament, the quality of the original recording in terms of mike setup, etc. can fully be had from PCM done properly at 4x rates (24/176.4 or 24/192), i.e. he has testified that it is audibly indistinguishable from the live feed in his studio. Also according to him, IIRC the same cannot be said about DSD, and, based on my own, personal experience, I have to perfectly agree with him.

 

That being said, one of the most plausible explanations as to why this is so IMO is the long established fact 1-bit quantizers, when properly dithered, are in constant overload. The resulting problems are distortion, noise modulation, and idle tones.

 

While it is true that choosing a higher order Sigma-Delta modulator shifts more noise out of the audible band, it is also true that there exists an optimum order when designing such a modulator, due to practical constraints. However, oversampling not only shifts even more noise out of the audible band, but also shifts it farther above the audible band, where it can be more effectively filtered.

 

On top of that, decimation increases accuracy by means of averaging, i.e. decimation does not in any way imply that redundant sampled values are merely being discarded. So in modern audio ADCs outputting PCM, there is usually a more complex combination of the digital filter and several other optimizations. The idea that properly done DSD is more accurate or more close to the original analog input signal than properly done PCM is a myth.

 

Thanks for a post which I found very informative.

[Boris75]

You will get different opinions from DSD advocates such as Cookie Marenco of Blue Coast Records

 

Sure, she records to analog tape and then digitises the result on DSD: in such a setup, given that DSD128 has higher resolution than analog tape, DSD128 captures everything that's on the tape (as does 24/96 PCM). DSD64 fully captures the dynamic range of analog tape but, if the tape includes some signal above 22kHz, it will be drowned in DSD64 quantization noise.

[Hiro]

if the tape includes some signal above 22kHz, it will be drowned in DSD64 quantization noise.

 

DSD64 has typically around 90dB of dynamic range up to 50kHz. Now, what was the dynamic range of tape again?

 

The new VAD DSD Player has frequency response up to 100kHz and dynamic range of 110dB with DSD64.

 

Of course, today, no one is stoping record labels from using the higher DSD128 rate, as the playback capability of this DSD rate is pretty much ubiquitous now.

[Paul R]

That happens when the modulator is NOT properly dithered, and it doesn't matter if it's 1-bit or any 1+N -bit. Especially with low-order modulators.

 

I have been very careful to design my modulators and the dithering scheme and I have been analyzing and measuring those a lot, as well has bunch of hardware multi-bit modulators inside DAC chips.

 

Most of the time DAC chips perform badly when converting from PCM input because 1) they have inadequate oversampling filters, 2) they use low-order modulators.

 

 

 

Too bad most of those ADC decimation filters are also bad, offering only something like 100 - 120 dB stop-band attenuation. I prefer to use >200 dB...

 

Decimation is just a useless step because eventually you'll need to oversample and modulate again for the playback. Plus you get time domain effects of the decimation filter.

 

 

 

PCM won't be "properly" done as long as it happens inside a tiny $10 chip. Very likely not the DSD either...

 

Well said, and very accurate. -Paul