Multi-bit DSD versus PCM

[Blizzard]

Well, my focus was on what happens when you add digital processing in the middle as every system should. If the bits are taken out of DSD domain, then calling the stream still DSD is just wrong.

I'll try to find where this is explained. There's probably a dozen threads with bits and pieces everywhere.

 

[Al M.]

Yes some are slower to catch on than others I don't think theres any dispute from anyone that quad DSD isn't superior. What their end of the story is the most popular formats can be done better (PCM) using their approach. But they are also not using third party software based SDM/SRC when making this claim either. This is assuming a regular native format is sent to the DAC as is. And once again this is based chips from yesteryear as well.

dCS is also still firmly in the PCM territory, and they have always designed their own chips and algorithms. I'll call their approach yesteryear only once I hear that the NADAC beats a dCS Rossini. I'll believe it when I hear it, but honestly I doubt it. Yet we will see. Until then, calling something yesteryear or not is premature. If DSD is your favorite thing, fine, but you haven't heard dCS gear either. So it's best to withhold judgment, don't you think?

 

[Blizzard]

Very very old articles:

• From Ayre: http://www.ayre.com/insights_dsdvspcm.htm

• Octuple-rate DSD: https://en.wikipedia.org/wiki/Direct_Stream_Digital

The results of that Ayre test will depend on the DAC used, and how the DAC handles the native formats sent to it.

 

[Blizzard]

dCS is also still firmly in the PCM territory, and they have always designed their own chips and algorithms. I'll call their approach yesteryear only once I hear that the NADAC beats a dCS Rossini. I'll believe it when I hear it, but honestly I doubt it. Yet we will see. Until then, calling something yesteryear or not is premature. If DSD is your favorite thing, fine, but you haven't heard dCS gear either. So it's best to withhold judgment, don't you think?

The flagship DCS stuff does exactly the same thing as HQplayer does. It resamples all PCM to DSD. Only it's limited to DSD 64, and uses far less powerful processors than Intel CPU's. Not only that they charge $20000 for their upsampling box to do it. It's targeted for audiences who think more money is better, and turn their noses up at software like HQplayer.

 

[NorthStar]

From what I'm reading what counts is the RECORDING.\

? http://www.positive-feedback.com/Issue26/dsd.htm

 

[Al M.]

The flagship DCS stuff does exactly the same thing as HQplayer does. It resamples all PCM to DSD. Only it's limited to DSD 128,

Source of your info please?

I know that PCM --> DSD is an option in dCS gear, but I had thought that their standard fare is straight PCM from PCM source.

 

[amirm]

The flagship DCS stuff does exactly the same thing as HQplayer does. It resamples all PCM to DSD.

You mean that by this they say on the web site? "In common with the award-winning Scarlatti, Paganini and Puccini ranges, Debussy uses our proprietary dCS Ring DAC™, which oversamples all incoming data to 5-bits at 2.822 or 3.07MS/s."

If so this is the perversion of term DSD I am talking about. What they are doing is upsampling the PCM to 2.8 Mhz which in the process reduces the bit depth to 5 bits. It is not DSD which was the intermediate bitstream of a 1-bit sigma-delta converter.

 

[Al M.]

You mean that by this they say on the web site? "In common with the award-winning Scarlatti, Paganini and Puccini ranges, Debussy uses our proprietary dCS Ring DAC™, which oversamples all incoming data to 5-bits at 2.822 or 3.07MS/s."

If so this is the perversion of term DSD I am talking about. What they are doing is upsampling the PCM to 2.8 Mhz which in the process reduces the bit depth to 5 bits. It is not DSD which was the intermediate bitstream of a 1-bit sigma-delta converter.

That's what I thought as well.

 

[Blizzard]

From what I'm reading what counts is the RECORDING.\

? http://www.positive-feedback.com/Issue26/dsd.htm

Of course the recording is #1. Music should be good as well . But even if just the good old redbook is used at the studio, lots of benefits can be had resampling it to DSD. This reason for this is because the DAC chips do this to redbook internally anyways. By doing it before the DAC chip using far superior algorithms, the end result is far superior sound. You will find that the debate always ends on this topic after people actually try it.

 

[NorthStar]

Joe Whip:

"I heard from one manufacturer that he wanted to sell his product for about $30,000 which is a ton of money. However, his Asian distributors told him he wouldn't sell any at that price and needed to price it in the $70,000 range which he did. Guess what, he had brisk sales. This kind of stuff really gets my goat."

___________

It's not the first time that I'm reading real stories like this...even here @ WBF there were other similar in that exact same target.

So, if dCS and their proprietary Ring DAC and Trinity DAC with BB PCM-1704 stacked DACS (8 per channel) in a unique special configuration...LIANOTEC architecture, from $30,000 (Trinity) to $90,000 (dCS multiple boxes) can be be compared with the HQplayer (Quad DSD) I'm putting my two feet anchored firmly on terra firma, and I'm listening attentively.
Now, that would be an interesting double blind test (triple).

Mike, how much?

 

[Blizzard]

You mean that by this they say on the web site? "In common with the award-winning Scarlatti, Paganini and Puccini ranges, Debussy uses our proprietary dCS Ring DAC™, which oversamples all incoming data to 5-bits at 2.822 or 3.07MS/s."

If so this is the perversion of term DSD I am talking about. What they are doing is upsampling the PCM to 2.8 Mhz which in the process reduces the bit depth to 5 bits. It is not DSD which was the intermediate bitstream of a 1-bit sigma-delta converter.

Jussi calls it SDM. It does what SDM DAC chips do when the PCM data is received. And optionally, it does a multibit conversion to the SDM so the DSP can be applied. In order for step 2 to happen, SDMdirect must not be checked in HQplayer Diff/DSDiff settings. SDMdirect, bypasses the multibit conversion step.

So not exactly the same. And Hqplayer can go all the way up to DSD 1024 with the upsampling if the hardware can handle it.

 

[Blizzard]

Joe Whip:

"I heard from one manufacturer that he wanted to sell his product for about $30,000 which is a ton of money. However, his Asian distributors told him he wouldn't sell any at that price and needed to price it in the $70,000 range which he did. Guess what, he had brisk sales. This kind of stuff really gets my goat."

___________

It's not the first time that I'm reading real stories like this...even here @ WBF there were other similar in that exact same target.

So, if dCS and their proprietary Ring DAC and Trinity DAC with BB PCM-1704 stacked DACS (8 per channel) in a unique special configuration...LIANOTEC architecture, from $30,000 (Trinity) to $90,000 (dCS multiple boxes) can be be compared with the HQplayer (Quad DSD) I'm putting my two feet anchored firmly on terra firma, and I'm listening attentively.
Now, that would be an interesting double blind test (triple).

Mike, how much?

The SDM/SRC is only 1 part of the pie Bob. The rest of it must be done good as well. But you don't need to spend that kind of coin that's for sure. These guys come up with elaborate ways to make their products very expensive.

 

[amirm]

Jussi calls it SDM. It does what SDM DAC chips do when the PCM data is received. And optionally, it does a multibit conversion to the SDM so the DSP can be applied. In order for step 2 to happen, SDMdirect must not be checked in HQplayer Diff/DSDiff settings. SDMdirect, bypasses the multibit conversion step.

There is no reason to rename anything. PCM allows any sample rate or bit depth. It is of no need of new name just because someone uses it at non-standard bit depths like 5 bits.

 

[Blizzard]

That's what I thought as well.

You mean this:

"Operating as a digital-to-digital converter, Vivaldi Upsampler accesses music from any digital source and converts the audio from its native sample rate to either high resolution DXD (24-bit data at 352.8 or 384kS/s), DSD (1-bit data at 2.822MS/s) "

http://www.dcsltd.co.uk/products/vivaldi-upsampler/

What I read was this:


"dCS is also still firmly in the PCM territory"

 

[Blizzard]

There is no reason to rename anything. PCM allows any sample rate or bit depth. It is of no need of new name just because someone uses it at non-standard bit depths like 5 bits.

Read my last post about the dCS Vivaldi upsampler. It works like this only can go up to DSD 1024, rather than limited to DSD 64, and has way more modulators and filters to choose from.

But Jussi names it SDM, because it's exactly the same thing all SDM chips do to PCM once they go into the SDM modulator section of the chip. If this is preformed prior to making it to this section of the chip, the data is passed through the chip unprocessed by the inferior SDM algorithms in the chip.

 

[amirm]

Read my last post about the dCS Vivaldi upsampler. It works like this only can go up to DSD 1024, rather than limited to DSD 64, and has way more modulators and filters to choose from.

I read it. It says it has three (3) modes of operation. PCM pass through. PCM upconverted to DXD, i.e. 24 bit at 352.8 or 384 Khz, or 1-bit DSD. Only one of the three modes is DSD.

 

[Blizzard]

I read it. It says it has three (3) modes of operation. PCM pass through. PCM upconverted to DXD, i.e. 24 bit at 352.8 or 384 Khz, or 1-bit DSD. Only one of the three modes is DSD.

Yes well HQplayer can convert to any PCM format up to 32/768 or any DSD format up to 1024 and vice versa. So similar operation, only the dCS is much more limited with sample rate choices, modulators, and filters. Also much much weaker processors. They use FPGA's rather than quad core Intel processors. The very best algorithms require very powerful processors. Jussi has filters and modulators prepared for CPU's that aren't even out yet. Today's CPU's will choke out trying them. His plan is to simply keep up with Moore's law, and release them as processors get more powerful.

He could build a box like the Vivaldi Upsampler too. But only his crappiest algorithms would work on it simply because the FPGA's are too weak. Why do you think the vivaldi upsampler maxes out at DSD 64? He told me about a new FPGA that could handle them but the chip cost alone was around $20000. And still less powerful than a $200 Intel processor. If dCS used that chip it would likely be a $200000 box retail.

 

[esldude]

I wished I were more convinced all of this sounds better. Or even perceptibly different.

I have used HQplayer as a trial. It does do excellent filtering. Even using it with a PCM DAC that tops out at 192khz, his filtering gave measurably superior results at the analog out when playing redbook files. The areas of superiority were small, but real. I don't have a DSD option, but I don't doubt it would perform excellently as he says.

Despite all of that I didn't hear a difference. Even old plain PCM can be superbly good. Convince me we aren't gilding the lily here.

 

[BlueFox]

These discussions are no different than discussions over how many angels can fit on the head of a pin.

 

[Blizzard]

Here's a great example to clarify things. This is the block diagram of the new AKM AK4497:



You can see on the left where the PCM data interface and the DSD data interface is.

Let's follow the data path of the PCM first:

First it goes into a "soft mute" this is to eliminate pops and clicks when the internal registers are accessed. From there it goes into the oversampling interpolator where the resource constrained filters are applied. Then to the modulator. Inside the modulator, the data is converted into multibit SDM. It uses the best modulators they could come up with based on the limited power of such a tiny little chip. This is where the digital volume control can be applied. then from there to a filter, and then the analog outs.

Now lets look at what happens when you send it DSD. With DSD there's 2 different options. You can send it through the DSD filter, soft mute, multibit SDM modulator if you want. Or you can bypass all of that stuff. Option 2 completely bypasses all of the resourced constrained SRC/SDM inside the chip altogether, as well as the multibit conversion. It's by far the purest path in the chip. So if you use HQplayer to handle the SDM/SRC with it's far superior SDM/SRC algorithms, you can send all of the data through option 2 of the chip. There's no better way to do it.

Now if you have a Sabre chip, or any other SDM chip (or use option 1 for DSD on the AKM), and send it DSD, it's also better. This is because if you do the SDM/SRC in Hqplayer, it still runs through those blocks, but it doesn't use the modulators because it's already done. It doesn't use the PCM filters either that aren't that great compared to HQplayer because you up sampled to the highest format the chip can process in HQplayer already. It can't upsample higher, so it doesn't do it. So the resource constrained filters are left alone.

However it still does the multibit conversion for the volume control. So if this it your only means of volume control for your system, this is the path you must take. If DSD option 2 was used, you must have an analog volume control somewhere in your system. So now the big choice is what's the lesser of the 2 compromises?

Just remember, by the time the data leaves the modulator, whether it started out PCM, or DSD, it's now the same thing. Only the PCM went through far more processing with inferior filter/modulator algorithms than the DSD did.