ANOTHER Example of Why I HATE DSD and Why Customers Who Bought Sony's Boloney Are So Annoying

[Miska]

Miska ... I'm not 100% sure it is as it appears, but have you come across Leema Accoustics power amps which (they say) are variable gain - though that might just turn out to be a variable attenuation on the input!

 

One of my favorite hardware manufacturers - Accuphase also has selectable gain in their power amps. Four options, 0 dB, -3 dB, -6 dB and -12 dB. Not a huge range, but helps matching volume control range.

 

Their AAVA volume control is also pretty good.

[Miska]

Using resistor attenuation there will result in a minor degradation, and possibly why the lower output impedance of a good preamp makes a worthwhile improvement for many. That's also why you don't normally see the lower -3dB point of an amplifier set by the capacitance value of the input capacitor these days

You are more likely to see something like a 22uF Blackgate ( if any left!) or a quality electro of a similar or higher value..

 

It's still better than added distortion and noise of another entire device on the path with more cabling, connectors, etc.

 

Series input capacitors? No way, I have servos. Serves two purposes, negates any input offset plus doesn't require any internal offset adjustments either.

 

Why would pre-amp have any better stuff inside than a DAC? Or other way, why would DAC designer decide to have a bad design or worse than for a pre amp, especially when designing DAC to be connected straight to a power amp?

 

 

The series resistor is fitted to ensure stability of the opamp into the capacitive load of a cable.

 

50 ohm series resistor with special capacitance and output impedance compensation is better choice. I'm not at all sure that all pre-amps are properly designed in this respect either.

 

The LME49710 and LME49720 are particularly touchy as regards working into even a short cable load.

 

I have not used those yet anywhere. But selecting right opamp for right function is important part of the design. I feel sick when I read some people randomly swapping opamps into a DAC without understanding requirements for a particular function. In my designs it would even likely cause burned opamp (the two extra pins in a single op-amp have varying uses) and for some opamps it won't even fit, because not all use the standard pin layout (single op-amp in a 16-pin SOIC case).

[Miska]

You know I agree with your approach (I'm an NOS PCM1704 guy feeding it 384kHz), and while yes, modern S-D DAC chips are cheap, why do we have to settle for cheap? I wish there were more firms (like MSB and TotalDAC) producing discrete R2R ladder DACs with laser-trimmed resistors--and offering them as modules.

 

I'm fine with that too, but from technical perspective it goes this route...

1) You fine tune R2R ladders to max

2) You want to oversample at high rate to get best possible analog reconstruction with a simple analog filter

3) Settling time vs precision becomes problem and you hit the realization edges with gigaohm at one end and milliohm at the other

4) You realize that with noise shaping you don't need as many bits at such high oversampling rate anyway

5) You start dropping off bits and increasing frequency as you go, because you can, and performance keeps increasing

6) You end up with very high frequency and just one bit (practically PWM)

7) You realize most of the performance is tailorable as digital domain algorithms

8) DAC becomes almost pure software with very simple hardware

9) You can finally put all hardware effort into making very good analog design

 

But to me that is like running the race and stopping just before the finish line: Don't give your lovely DSD512 (s/w modulated) signal to an ordinary DAC chip to have it poorly filtered and handed off to some compromised output stage. Just amplify it a bit with video amps and use a low-pass analog filter--to have a DAC-less DAC. See SONORE Rendu PureDSD DAC (it just does not take DSD512 and is DLNA-only at the moment).

 

That's what I'm working on now. I've got my own CS4398 Direct DSD & analog filter for couple of years now, very good for DSD128. But now I'm going for discrete implementation and DSD512. But just passing bitstream to an analog filter doesn't give best possible performance, it needs more.

 

Since I'm constantly busy on software, it takes time to get something ready on hardware side. I don't mind if something suitable appears on the market, would save a lot of time & effort.

 

I just need to do some hardware stuff every now and then to stay somewhat sane...

[Miska]

I am sure that whatever you are working on will be great. I just don't like it when manufacturers disparage other approaches so as to promote their own design.

 

I'm not a hardware manufacturer. I have no intent to sell any of the hardware I design or build. I design and build audio hardware for my own proof-of-concept purposes and because I've been DIY audio guy for 25+ years. (since there's always a chicken-egg problem for new technologies, something needs to be done to test software when hardware is not yet available on the market)

 

I've designed commercial hardware for non-hifi use (measurement, etc), which is completely out of context for this forum and I cannot even talk about that stuff anyway.

 

So count me as a DIY guy when I'm talking about hardware.

 

I can say from my experience that it is very much possible to get unbelievably great sound at very low speeds. That method works well in Vincent's design. Certainly, it's not your approach. As to your synopsis of all low speed DACs; II'll take that with a huge grain of salt. :-]

 

What is low speed DAC here? For example MSB DAC is 1.5/3 MHz which is not low speed, it is very high speed for a multi-bit DAC. (for comparison, PCM1704 chip can do 768 kHz)

 

I don't get to subjective evaluations because there's no end on that discussion. But I have not seen a DAC that runs actual conversion stage at frequencies below 200 kHz and still offer competitive measured performance compared to delta-sigma designs.

 

HQPlayer supports PCM upsampling up to 1.536 MHz 32-bit with number of dithers and noise shapers. It also supports delta-sigma modulated output up to 24.576 MHz 1-bit. And all the processing is available for both, there is also specific support for older 18-, 20- and 22-bit DACs to fully utilize those without truncation. So I'm fully in both PCM and DSD camps. Base line is that I seek for ways to achieve best possible performance at most reasonable price. Currently I feel that I can get closest with DSD, but I reserve right to change my mind at any point in time...

[Miska]

By the way, that 9-step list is pretty much how current DAC chip evolution has gone. At least following companies seem to have concluded pretty much same way since they now produce delta-sigma DAC chips, even though many had multi-bit ladder designs in past:

- Burr-Brown/TI

- Analog Devices

- Philips/NXP

- Crystal/Cirrus Logic

- AKM

- Wolfson

- ESS

- Seiko NPC

- Niigata Seimitsu

 

Would all these be wrong, of course a possibility, but there is probably some commonly shared view behind their choices anyway?

 

P.S. And to be exact, BB/TI uses hybrid design in their best performing chips, 6-bit ladder combined with delta-sigma handling rest 18-bits (18 + 6 = 24) with actual analog stage operating with 66 levels.

[Miska]

What are the universally accepted DAC measurements that 100% positively correlate with listening pleasure? If those measurements cannot be shown to be relevant to listening enjoyment, then why measure? A DAC manufacturer that measures AND listens may be one who has accepted that he can't know all of the best measurements that strongly correlate with listening fun.

 

I've been measuring all my DIY hardware and also all the purchased hardware I've ever had. I believe I have some kind of idea of correlation between what I hear and what I see in measurements. And also kind of estimate of how certain type schematic will sound. Same for DSP algorithms too, I've been analyzing, simulating, tuning and listening DSP algorithms. Of course I still have a lot to learn, if I ever feel I've learned everything I will most likely go for something completely different and stop tweaking audio.

 

I always both measure and listen, but I'm sort of engineering perfectionist in a way that I demand both perfect measurements and sound (listening). I've probably told many times about my ranking system when I buy audio hardware... First I use measurements to reduce the potential hardware to a small group (thank you JA and PM!), on which I then perform my listening tests to select the one that sounds best. Same method is used for designing DSP algorithms, first an analysis check that the technical performance meets expectations and then listening tests to evaluate sound. And this cycle is repeated over and over again.

 

 

but there's little evidence that even DAC linearity strongly correlates to listening enjoyment.

 

It is possible learn to hear all kinds of errors and once you know those, it is very painful to listen because it bothers you all the time...

 

I have my quest to reach most detailed and accurate sound that is pleasing to listen to. Others can then evaluate how well I've got there regarding DSP stuff. I also tell the commercially available hardware I use myself.

 

The B&K house curve has many cousins and they all have one thing in common: high frequency rolloff and boasted bass. I am not saying that everyone should start using target based DSP but when they do, the preferences are almost always non-flat targets. Why is that? I don't know.

 

I covered this topic in my post here, on DSP area to the thread regarding house curves. Optimal house curves depend on speaker radiation pattern and acoustic properties of the room, and also listening axis vs the speaker.

 

Jussi, even discrete SDM DACs are cheap to build. You only need 1 resistor and the precision of that resistor is irrelevant to the sound. That's the reason you see so many SDM DACs now days. Sure they sound great but they are not categorically superior to other great multi bit designs in terms of listening enjoyment.

 

Well, it is more than one resistor. From objective evaluation point of view, if you can get 20 dB better performance for 1/10th of the price I don't see too many negative sides on this. For evaluating listening ejoyment, I prefer using standardized MOS style scoring methods.

 

I'm not trying to say multi-bit ladder DACs are bad. And performance of those can be also improved by upsampling and noise shaping. But so far it is a road where price increases exponentially and performance logarithmically. I'm trying to reach the inverse, logarithmic price increase and exponential performance increase.

 

You can find some interesting listening test results for lossy codec implementations here:

http://en.wikipedia.org/wiki/Codec_listening_test#Results

 

For example:

http://listening-tests.freetzi.com/html/AAC_at_128kbps_public_listening_test_results.htm

[Miska]

Both my Class A preamp and 15W Ch. Class A P.A. are both fully DC coupled with servos. There are no capacitors in the signal path or the feedback networks of either. The front ends of both also use dual metal can , closely matched transistors,(LS313 and LS352) with front end balancing as posted in the DIY Audio Current Mirror thread of Nov.2008.

 

Sounds good. There used to be plenty of factory matched bipolar transistors and J-FETs, but nowadays it seems to be hard to find those. SSM2212/SSM2220 is still available and couple of newer MATxx models, especially MAT12 looks promising.

[Miska]

7) Sony played one of their typical dirty tricks when they announced that DSD would play UP TO 100 kHz with a dynamic range of UP TO 120 dB. This is complete and utter bullshit, as the S/N ratio of DSD at 100 kHz is negative. If you look at a reasonable bandwidth of say 20 kHz to 100 kHz, there is MORE NOISE in the DSD system than there is total signal.

 

Oh really? Here's 100k tone encoded into DSD64:

 

11) So when people started comparing the SACDs against either CD's or DVD-Audio discs. the combination of FAR BETTER hardware, along with FAR BETTER MASTERING, along with the (virtual) elimination of brickwall filters, most people said, "Aha! Sony is right! DSD is a better format than PCM!"

 

Seems like you love the sound of aliasing in PCM. You would need input frequencies exceeding 1.4 MHz to make DSD alias... Not so hard to get non-brickwalled quad rate PCM to alias with microphones like this:

SANKEN MICROPHONE CO .,LTD. | Product [ CO-100K ]

 

and since the DVD-Audio committee was SO STUPID as to require a completely different player than a regular DVD player, it was also doomed to failure.

 

DVD-Audio died, because the DVD committee didn't understand how to make a proper content protection, so breaking CSS was extremely easy. Unlike the SACD protection.

 

In today's online business, way of content protection is to watermark the content data with buyer's information. So whenever pirated content circulates, it is easy to know where it began.

 

14) The story would have ended there except for Gordon Rankin. He had been on parts of the USB committees over the years, and he made sure that USB had the ability to make a good audio data transfer link. So when he released his "Streamlength" asynchronous isochronous USB firmware, all of a sudden any computer made in the last ten years could serve as a transport of higher performance than ANY conventional S/PDIF transport EVER MADE!

 

Lovely marketing speech, but now you are forgetting all the pro-audio gear that had asynchronous USB audio years before hifi gear. Let's not have a tunnel vision here.

 

15) Now we have to split the story somewhat. First we will look at multi-channel PCM. Whilt it is TRIVIAL to record and make multi-channel PCM files for USB DACs, there is virtually no software available, very few software players that will handle it, and very few USB DACs are made with more than two channels. Essentially it is a solution waiting for a problem to solve.

 

Again you seem to have never looked at pro-audio side. Already 10+ years ago there was no problem having 64 channel recording and playback with a computer.

 

I purchased my M-Audio Delta1010 in year 2000. It was limited to simultaneous 10 channels in (8-analog + S/PDIF) and 10 channels out (8-analog + S/PDIF) at 96/24, but that was pretty good at that time for inexpensive hardware. By the way, that hardware is still being made and available. I used it a lot for recording and playback and wrote quite a lot of software for using it for all kinds of thing on Linux.

 

So we will leave that alone, except to say that if one thought it were a viable market, it would be trivially easy to make a 16-channel 192/24 USB playback system. But nobody really cares except a handfull of people that generally just purchase Blu-ray discs of live audio concerts.

 

Quite a lot of pro's buy those things, of course luckily there's good hardware available if you are into PCM:

RME: Fireface UFX

 

But most still use Firewire because it can perform better and is more reliable.

 

I first heard about it in 2008 (I think, I can't remember any more). I was seated next to Gus Skinas, who had been part of the Sony SACD team. His role was to be the liaison between the recording studios and the technical people at Sony who would loan out the hardware required to make an SACD. When he told me about it, I talked to Gordon Ranking about it and we said that it would be trivial to packetize the DSD stream so that it looked like PCM.

 

At that time, I already had such scheme for transferring DSD over PCM (with FourCC marker 'DSD ') because I needed it for my prototyping, I also got emails from companies asking for such feature. But I usually recommended them to just provide ASIO drivers with DSD support, because it is not such a hack and supported both record and playback. ASIO had native DSD support already early, I think at least 2006 or earlier. I cannot really remember and I don't have a good version history.

 

At that time I also designed my A/D/A converter that has analog input with two ADC chips in parallel, one running at DSD128 and another one running at 192/24. It also has a DAC chip for playing back either of those two contents.

 

But then we realized it was a fool's errand because the only source of software was to (illegally in this country) rip an SACD with one of the rare specific models of PlayStation 3s. So we said, "Screw it."

 

I didn't worry about the content, because people could record DSD and for me DSD was a way to turn a modern DAC chip into "NOS" mode and run all the digital processing in a computer. Around those times there were also Korg MR-1000/MR-2000 recorders that were very good for archiving vinyls at DSD128 and lot of people were using those recorders for that purpose. But those recorders were inconvenient for normal playback, mostly because of limited storage space.

 

Here is one report from 2007, from Stereophile site regarding vinyl archiving with a Korg recorder:

http://www.stereophile.com/content/korg-mr1000-used-archive-vinyl-1

 

The problem is that now Sony has to talk out of both sides of their mouth at once. They painted themselves into a LITTLE TEENSY CORNER because they said that one of the prime advantages of a one-bit system was that was always inherently linear. But when people found out the trught that probably less that 0.001% of all DSD recordings actually were transferred into PCM and then back to DSD, they look pretty damn stupid.

 

Now you don't seem to understand this part. PCM is also inherently linear as long as you stay in digital domain. Problem is that at the actual conversion stage, only 1-bit is inherently linear while it is impossible to make a perfect multi-bit converter because proportional ratios between ladder steps are never exactly what they are ought to be, due to component tolerances.

 

18) But the truth is that there are still some HUGE problems with DSD, especially with regards to out of ban noise

 

It is not a problem, it is only a problem because you are trying to compare Nyquist-sampling system to a non-Nyquist sampling system. If you disregard Nyquist frequency and look at PCM at equivalent bandwidth, for example from 0 - 1 MHz it also has out of band noise issues, especially if you don't use brickwall filters. PCM especially has a HUGE problem of this out-of-band noise (images) being directly correlated with the input signal!

 

When I measure a DAC I use at least 5 MHz bandwidth to see what it actually pushes out.

 

19) We have recently introduced a two channel A/D converter that will output both PCM and DSD. But this converter has a few tricks up its sleeves. Specifically we have taken a page from the "What is so great about DSD" manual and applied it to PCM. And it turns out that with quad rate PCM you can get ALL OF THE SONIC ADVANTAGES OF DSD WITH NONE OF THE PRACTICAL DISADVANTAGES.

 

UMM, lemme guess... It will have heavy aliasing if I input 2V 100 kHz tone to it while recording quad rate PCM.

 

20) So the bottom line is NOT that you are an idiot if you like DSD. DSD can sound wonderful. But what I am saying is THERE IS NOTHING MAGIC ABOUT DSD. WE CAN GET ALL OF THE GOOD THINGS ABOUT DSD IN A HIGH SAMPLE RATE PCM RECORDING ALSO!

 

How high are you ready to go? At least 352.8 kHz minimum and preferably 1.536 MHz, at 24-bit. Oh, wait, now the file downloads become impractically large, even for modern network connections.

 

DSD has very good bitrate efficiency vs sound quality.

[Miska]

At DSD256 (already available), we can do something like this

 

or alternatively something like this

 

At DSD512 (also already available as DAC) it can look like this for example:

 

Guess how it will look like with DSD1024?

 

And the noise floor at low frequencies in these images is limited by source and destination material used to make this plots. Not by the DSD...

[Miska]

And here is how I understand Charles thinks 100 kHz sine recorded at quad rate PCM should look like:

 

IMO, that alias is much more nasty than non-correlated hiss at the same frequency.

 

I would at least expect to be able to record to 100 kHz with a mic that has 100 kHz response. DSD256 works pretty well for that purpose.

[Miska]

The fact that the DSD sounded better to him could easily be due to the DSD files in question simply being better recordings, and have NOTHING to do with the relative sound of the 2 formats.

 

However, one could conclude that at least DSD is not the limiting factor for sound quality either, although "PCM proponents" are trying to argue that DSD is not good for anything and that PCM clearly outperforms it.

 

My opinion has been that one performance limiting factor of both ADC and DAC has been the DSP section. Since DSD is closest to bypass of those sections, it is one way to lift some of the limitations.

 

So what if DSD difficult format for mixing and mastering? They do it as work and get paid for it, so they should be the ones going over all the hoops. And in any case, it's all done once and that's it. But all the people who listen the content should get very good performance from very reasonably priced hardware, so the playback hardware reproducing the content shouldn't have to go through all the hoops every time.

 

RedBook was very easy format for record companies, but it took 30 years to get the playback performance at reasonable level and it takes enourmous amount of DSP power to do it. New content formats should simplify the playback end and move as much complexity as possible to the production stage. Hires PCM is definitely not that thing, because it just increases DSP demands at the playback side. DSD may be it.

 

Or maybe 8-bit unary coded (9-level) delta-sigma at 5.6 - 24.576 MHz as I proposed earlier. This is what I'm going for anyway.

 

Meanwhile, IMO, DSD is a good way to go. Or if you prefer PCM, then DXD.

[Miska]

Charles is frustrated, because there is no need for DSD to make superior recordings, what is required is 4x PCM, with a well designed digital filter.

 

It is just not possible to have a good non-brickwall filter at 4x PCM rate. You need to get aliases out of the data so that what ever you input to DF, there are no detectable aliases included. For proper sound, you also need to be able to record all harmonics that the instruments really provide, until they disappear in the background noise.

 

Even if you choose to begin cutting out at 20 kHz, you have only two octaves above before Nyquist. That means the digital filter needs to have at least 72 dB/octave attenuation. That still sounds like a brickwall, doesn't it?

 

While with DSD64 you have six octaves to spare for similar frequency response so 24 dB/octave is enough, that would be a fourth order filter which is completely realizable already in analog domain. But since the noise floor raises as function of frequency, you don't need even that much, because there's no 144 dB SNR anymore at the Nyquist. So second order analog filter is enough...

 

Having a second order digital anti-alias filter for 4x PCM is definitely not enough. And not fourth order either.

 

For DXD, 48 dB/octave is enough which is eight order filter, just within a reasonable analog filter design and on the edge of being tolerable. But you would still have just 20 kHz pass-band, so you would get ringing and be altering properties of output from a 100 kHz mic.

 

The only thing "magical" about DSD sound, when one gets around the DF issue, is likely the presence of large amounts of HF noise, and how this interacts with the rest of the system. I suspect, that this noise is often (but not always) responsible for the "sweet" "soft" sound of DSD often present in many systems.

 

And you don't need to consider similar noise generated by DAC chip's delta-sigma modulator, because you don't see it in digital domain prior to the DAC? Plus all the digital images created by poor oversampling? What you see doesn't exist?

[Miska]

To compare apples to apples, here is digital domain simulation of how output from a typical "flagship DAC" chip delta sigma-modulator output looks like. This is fifth order SAH delta-sigma modulator to 5.6 MHz with dithered 6-bit output. Source is 100 kHz tone using 352.8/32 PCM (output format of such DAC chips' 8x interpolation digital filter).

 

No out of band noise? Drastically different from a DSD? Really?

 

 

 

I can only say ugly...

[Miska]

But can a normal listener hear it

 

Not the noise, most of it is removed by combination of the actual D/A section and following analog reconstruction filter stage. Since those digital images are directly correlated to input signal and have higher level, those may cause adverse side effects to the audio band in form of intermodulation products. (intermodulation products of white noise are still white noise, they just increase the base noise floor)

 

Design properties of the modulator affect the sound, it is one source of the DAC's sonic signature. Martin Mallinson also said this in his presentation.

 

In my opinion sonic properties of a DAC originate in following order per domain. Both domains are equally important.

 

Digital:

1) Delta-Sigma modulator

2) Digital up-/oversampling filter

 

Analog:

1) Design of the actual D/A section

2) Analog reconstruction filter

3) I/V section

[Miska]

Miska: you continually make this claim, that PCM reproduced by the typical DS DAC these days (and there are still plenty of R2R DACs to choose from if one really wants that option) has the same noise spectrum as DSD. Now this may be true with what you are doing personally (which no one else has access to), but, al it takes is a quick trip over to stereophile.com, and to check the measured noise spectrum of various when DACs when playing back DSD vs PCM. John Atkinson's measurements clearly show the much greater noise content, much closer to the audio band with the same DAC playing DSD vs PCM.

 

There are couple of reasons.

 

I don't know what you are referring to in terms of what I'm doing personally?

 

1) Those DSD measurements are made using DSD64 that runs at 2.8 MHz. PCM DACs commonly use 5.6/6.1 MHz, or 11.2896/12.288 MHz. So the noise is shifted up proportionally same way as for DSD128 and DSD256 that correspond to those frequencies, so it begins 2x or 4x higher. Sabre runs at rate equivalent of DSD1024 and it's D/A stage (analog filter) cuts out most of it's OOB-noise. As it should.

 

2) Design of the D/A section defines where it begins to cut into the noise. Most have been designed for higher frequencies, but TI chips allow four options to reconfigure the conversion section for DSD. CS4398 is optimal at DSD128.

 

3) If you look at measurement results in 4x PCM mode, the frequency response typically starts rolling off quite early vs Nyquist - to suitably cut out the raising noise slope. Shouldn't 4x PCM frequency response go flat to 88.2/96 kHz?

 

Most measurements conveniently stop at 100 kHz or 200 kHz which is too low to actually see what is going on with DACs.

 

There is no magic involved.

1-bit has dynamic range of:

20*log10(2¹) = 6.0206 dB

6-bit has dynamic range of:

20*log10(2⁶) = 36.124 dB

 

So the overall noise level difference is just 30 dB, but at the highest frequencies both hit 0 dB at noise level due to noise shaping.

 

This is, of course, for SD DACs, which I admit does cover the majority of them, including my own. But if one really wants it, there are still options for ladder style PCM DACs. Off the top of my head I can think of a few: Naim, Resolution Audio Cantana, Total DAC, MSB, Phasure, Ayon...

 

The new Naim DAC-V1 is delta-sigma DAC like their CD and network players too.

 

How big market share do you R2R has, if you cover all Blu-ray players, DVD-players, TV's, AVRs...

 

But you can check out my blog posts comparing two early 90's DACs, one is 1-bit 64fs and another one is R2R ladder. With measurements. Ladders DAC's output becomes dirty right above the Nyquist frequency it runs at.

Here is the CS4328 playing back 1 kHz tone, it is 1-bit 64x oversampling 5th order delta-sigma DAC - thus practically it converts 44.1/16 input to DSD. No large OOB noise here? The conversion stage is just optimized for the particular input rate and 22.05 kHz audio band.

 

Here is wideband spectrum of CS4398 playing back dithered digital silence at 192/24 PCM, measured from DAC's analog outputs (includes analog reconstruction filters). You can see the familiar noise bump. Internally it runs at 6.1 MHz here.

 

Here is DXD frequency sweep played back as DSD128 through Sabre DAC:

 

Here is DXD frequency sweep played back as 192/24 PCM through Sabre DAC:

 

Huge difference in out of band noise?

[Miska]

The above is a matter of opinion. Clearly, your opinion differs from that of Charles Hansen. Nothing wrong with that. But until you produce commercially available products, which win awards, and are greatly respected as outstanding performers, I know who I am going to believe in.

 

Well, I don't see that matching the QA-9 either. Maybe I'll get one for evaluation and post some measurement results. But I wasn't at all impressed on the Stereophile measurement results either.

 

I don't do hardware, I do software. So if your opinion depends on hardware solutions, then I guess I cannot affect it.

 

I have my own opinions and I like to discuss technical views, but I'm not trying to compete about "awards" and other such BS I'm not interested in first place. I like to stay purely technical.

[Miska]

The real world measurements of actual DACs one can buy are there at stereophile.com for anyone to view.

 

Where are all the wideband measurements up to 5 MHz that could actually show the out of band noise?

 

Not talking theory here, what happens with real world DACs which audiophiles purchase. The measurements show more HF noise at a lower frequency with DSD playback.

 

Yes, that's what I'm talking about too. But could you point to some particular measurements?

 

Yes, this test is with DSD 64 content, as that is what 99.9% of all available DSD recordings (SACD) are. What is possible in theory is not what I was referring to, I was referring to DACs (and disc players) which are available to go and purchase at your audio dealer.

 

We were talking about what happens inside DAC chip here, not about the content.

 

But I upsample all PCM content to DSD and many other do too. There's no reason to stick with DSD64 in those cases.

 

There is big difference to what happens in the real world, where most of the audiophiles live, and the products which they can purchase for playing back music, and what may, or not, happen in the world of Miska, where SD modulators are custom, DACs are custom, other hardware is custom, etc. What I am saying is that the results which you have are not available to the average audiophile, and as such are irrelevant (except theoretically).

 

What!? They are very much relevant. Measurement results are from real world DACs. Simulations are simulations, because otherwise you have to laser-cut the DAC chip open to hook into insides. And sometimes it is necessary to discuss theory and technology development too.

 

But mostly I talk about modulators in (1) DAC chips, (2) in HQPlayer. Both are available to everybody.

[Miska]

Some more real world measurements.

 

Here is RedBook frequency sweep through a DAC based on Burr-Brown DAC chip (their hybrid design):

(notice the delta-sigma noise bump at 500 kHz vs the images every 352.8 kHz caused by SAH from the PCM data, noise is non-correlated and the images are directly correlated)

 

Here is DXD frequency sweep through the same BB based DAC:

 

Here is RedBook frequency sweep through a DAC based on R2R Burr-Brown DAC chip and their (brickwall) DF chip:

[Miska]

But I think what REALLY chapped his britches was that he had been convinced (by someone!) that DSD was infinitely superior to (ugh!) nasty old PCM. And the fact remains that it is IMPOSSIBLE to change the volume of a DSD data stream unless you first convert it to PCM, adjust the volume (losing resolution along the way) and the remodulating it BACK to DSD (adding more noise to the signall).

 

This is wrong, you don't have to convert to PCM. You can keep it in SDM domain all the way. If you don't know how to do it, doesn't mean it doesn't exist.

 

If you double the SDM rate while doing it, the output can easily have 10 - 20 dB lower noise floor than the source material.

 

Same amount of noise is added in digital in both PCM and SDM cases, if you use same input and output resolutions.

[Miska]

This was confirmed to me back in the early 2000's by one of the Philips engineers. He said that by playing with the frequency response of the noise shaper that you could get the system to sound almost any way you wanted it o sound -- soft and sweet, hard and edgy, smooth and creamy, et cetera. So the noise shaping curve they used was deliberate selected to give a specify "tonality" to the process.

 

Sort of yes, large part of how a delta-sigma DAC sounds like originates from the modulator design properties. But DSD doesn't sound better because of the noise. PCM sounds worse than DSD because PCM has either 1) brickwall filter or 2) hefty amount of directly correlate digital images that cause intermodulation products.

 

This has been very clear for a long time already.

[Miska]

Can I ask which commercially available hardware you are impressed with? How about dcs Vivaldi or MSB Diamond?

 

I'm very hard to impress, I'm almost never impressed. But I try to pick up my highlights of things I currently don't have as example.

 

I was impressed by Devialet, because it had some original thinking and good performance. And also by NAD's direct digital amps. I'm impressed by Accuphase how they keep producing nicely performing good sounding amps with great build quality.

 

I like the technology of dCS, Playback Designs and Meitner/EMM. And maybe MSB too, but I don't know enough of it to really have an opinion. But I'm not impressed by the price tags, not available here and outside of my budget.

 

I am particularly impressed by Chord QuteHD especially if it gains DSD128 capability, because it combines original engineering, good performance and reasonable price tag.

 

And of course I'm happy with the equipment I have.

[Miska]

Do you think the volume control in dcs DAC is also done in SDM domain?

 

I don't know, maybe for DSD sources. For PCM sources it could be done in PCM or inside the modulator.

 

ESS and Cirrus Logic DAC chips have digital volume for DSD. I think AK4399 also has, but AKM doesn't tell anything how they do it.

[Miska]

Actually, SDM is not a domain either, is a method.

Correct is pulse-density or PCM 1 bit.

 

Where do you get that PCM there? It's not PCM in any way. Multi-bit SDM is not PCM either.

 

- If you feed 8-bit SDM to an 8-bit PCM DAC output is garbage.

- If you feed 8-bit PCM to an 8-bit SDM DAC output is garbage.

 

Check also the linked document:

1-bit A/D and D/A Converters

[Miska]

At the end of the day, DSD is PCM... The modality to obtain them is different, but resulting signal is 1 bit PCM. You can have PCM that is RZ encoded, true?

 

No it is not, in common multi-bit SDM formats bits have different meaning than bits have in PCM. Like I've said before, if for you any series of data (including this text encoded in UTF-8) means PCM I cannot help it.

 

You can transform any wordlength PCM into NRZ coding, but it won't play back on a DSD DAC.

 

In PCM you can have power of two number of levels; 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024, 2048...

In SDM you can have any number of levels; 2, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37...

 

P.S. Actually UTF-8 is closer to SDM than PCM...

[Miska]

In the same vein, Miska, can you tell us more about the method, range, and resolution of the s/w level control you offer for SDM in HQPlayer? I am really interested.

 

I'm much more happy to talk about "what" than "how"...

 

Range is practically unlimited, so the noise floor is limiting factor, not resolution.

 

And I'm all the time improving/adding algorithms when I figure out new ways of doing things, so things also improve over time, release by release.