Octave Music Don Grusin High Resolution Music Analysis (Video)

[Miska]

Even with that crude approach the noise peak is now at 270kHz and has comparable energy like typical content around 20kHz... low enough to not cause any problems downstream. The analog filter implementation is really tricky, though, with those horribly fast signal edges directly from the digital logic gate... that's why actual solutions can be much different and may show different final analog spectra.

And you can of course adjust the modulator noise slope so the it doesn't have that kind of sharp knee.

So it always makes sense to check the DSP + hardware combination in action. That's why I check with different bandwidths using different analyzers.

Same of course applies even if one uses stock DAC chip (without external DSP) with one's own analog stage.

 

[Miska]

I don't know that the 'required' 50 kHz analog LPF in question was properly considered part of the *DAC* itself, versus a downstream component of the hardware. Perhaps someone more savvy about SACD player design can tell us.

SACD specification defines minimum requirements for a low-pass filter. In that specification they have decided to choose 50 kHz as corner frequency where signal and noise cross each other and provides sufficient attenuation of the noise.

This frequency matches pretty well with reality if you record close miked drums/percussion or such. The exact point always depends on the content.

 

[Miska]

There is no way to perform transforms of any sort on a 1 bit format without it overflowing.

There is no way to perform transforms on any sort of 24 bit format without it overflowing.

Adding two 24-bit numbers produces 25-bit result. And multiplying two 24-bit numbers produces 48-bit result.

FYI, if you just turn on the DSP volume control in Roon, you lose DSD native playback. The format is that limiting. With my RME DAC also disabling its volume control, DSD is a pain in the neck to handle in its native form.

No, it disables DSP for DSD because it doesn't have power or algorithms to do it. It also turns off most DSP for 705.6/768k PCM rates for the same reason.

 

[JSmith]

In that specification they have decided to choose 50 kHz as corner frequency where signal and noise cross each other and provides sufficient attenuation of the noise.

... so the noise doesn't potentially damage speakers at over 100W output. Quite a liability, especially when often SACD users have pretty pricey systems.


JSmith

 

[Miska]

I wanted to know if Blue Coast, who are the other enthusiastic supporters of DSD do the same thing when they transform their files from DSD to PCM.

FAIK, Cookie mixes in analog. So the music is first recorded in DSD, then played to analog desk for mixing and then recorded again in DSD from the desk.

This is these days common workflow also for a lot of PCM content.

 

[Miska]

... so the noise doesn't potentially damage speakers at over 100W output. Quite a liability, especially when often SACD users have pretty pricey systems.

Yeah, no it won't, you will damage your hearing first.

I would be more worried about PCM image output from DACs damaging your speakers. It is many times higher in level than the DSD noise.

Not to even mention class-D amps that regularly spit out even more.

 

[Miska]

I did understand your question. As i noted, I already post Bluecost tracks. Here is the DSD128 again:

And again with arbitrary bandwidth and reconstruction filter.

Luckily that noise is again less than you get from many DACs as images with PCM inputs... And it is uncorrelated from the signal, unlike the fully correlated image distortion with PCM (thanks to those resource constrained DSPs in $10 chips).

 

[Blumlein 88]

FAIK, Cookie mixes in analog. So the music is first recorded in DSD, then played to analog desk for mixing and then recorded again in DSD from the desk.

This is these days common workflow also for a lot of PCM content.

Yes, this is how Cookie at BlueCoast does it. Two trips thru DSD land.

 

[Miska]

I mean it is ridiculous to read here, for class-D amp @amirm says there's "great attenuation" and yet it looks like this:

...and it would continue beyond that 1.2 MHz peak...

So DSD left-over noise that is tens of dB lower is some kind of problem.

When it comes to tweeters, I would be much more worried about that power amp. Or many DACs operating with PCM inputs.

And "high switching speed" at 600 kHz. Compared to 2.8 MHz of DSD64. Or 45 MHz of DSD1024... (yeah, we demoed an amp/speaker system switching at that 2.8 MHz at Munich many years ago already)

 

[Blumlein 88]

I mean it is ridiculous to read here, for class-D amp @amirm says there's "great attenuation" and yet it looks like this:

...and it would continue beyond that 1.2 MHz peak...

So DSD left-over noise that is tens of dB lower is some kind of problem.

When it comes to tweeters, I would be much more worried about that power amp. Or many DACs operating with PCM inputs.

Yeah I blew out several super ultra tweets with the 600 khz garbage in the output. Pretty costly, and even worse hard to monitor whether or not my super ultra tweets are working as it isn't easy to find 600 khz microphones to check on them. Plus at that frequency the air absorbs 60 db at 1 meter so with my 3 meter listening distance they are down 180 db and difficult to hear. Makes warranty claims overly hard to deal with. That is why I don't even try to monitor the ultra ultra tweets that cover the 1 megahertz and above range. They just blow sometimes and I don't worry about it.

 

[Miska]

Yeah I blew out several super ultra tweets with the 600 khz garbage in the output. Pretty costly, and even worse hard to monitor whether or not my super ultra tweets are working as it isn't easy to find 600 khz microphones to check on them. Makes warranty claims overly hard to deal with. That is why I don't even try to monitor the ultra ultra tweets that cover the 1 megahertz and above range. They just blow sometimes and I don't worry about it.

At least you can use the tweeter as hand-warmer on cold winter nights.

 

[firedog]

With PCM format you are not stuck with one or two solutions like you are with DSD. And no, software doesn't keep working if OS revisions break it.

And you are doing all this for what benefit? just some sonic improvement you haven't confirmed in a controlled test?

There's pretty much no SW that can be guaranteed to work forever. Even Google and Microsoft have removed functionality or killed SW that was used by many. I don't see that as a minus for HQP; it's no different than anything else on the market. And if it goes away I'll just use Roon. And if that goes away I'll use something else.

Ever hear of having fun?
I also use a tube pre-amp sometimes. Just for fun. It's not always a requirement to listen to textbook perfect reproduction for enjoyment.

DSD sounds a little different. And I've confirmed it in a blind test. But in the end it doesn't matter.
I'm not claiming it's superior or telling anyone else they should do it. Just like they shouldn't tell me it's stupid and a waste of time.

 

[Music1969]

I mean it is ridiculous to read here, for class-D amp @amirm says there's "great attenuation" and yet it looks like this:

...and it would continue beyond that 1.2 MHz peak...

So DSD left-over noise that is tens of dB lower is some kind of problem.

When it comes to tweeters, I would be much more worried about that power amp. Or many DACs operating with PCM inputs.

And "high switching speed" at 600 kHz. Compared to 2.8 MHz of DSD64. Or 45 MHz of DSD1024... (yeah, we demoed an amp/speaker system switching at that 2.8 MHz at Munich many years ago already)

Very solid argument here.

 

[KSTR]

Are you trying to say DSD file is just a string on 1 and 0 and no way to map a waveform, hence nothing can be done digitally? Unlike PCM can capture in frame and edit?

Exactly.
This is what a DSD stream, a single bit semi-randomly toggling at many MHz all the time (11.3MHz in this case), looks like:

The audio content is "buried" in that RF noise carrier, the noise is modulated with the audio content, a pulse density modulation (PDM). Because a bit is never repeated for more than two samples the vast majority of the energy is moved upward, near Nyquist frequency.

Direct editing on a 1-bit stream is limited to:
- cutting/trimming
- polarity inversion (flip bits)
- time inversion (bits backwards)

We always have to remember that a DSD stream is really just the captured modulated recording signal with the goal to keep the DAC hardware simple, in the most elemetary sense only an analog low-pass filter is needed to remove the carrier from the digital input signal. Editing was never within the scope of that total ADC-->DAC chain where the signal is tapped off after the ADC, stored (on file/SACD), given to the customer and the re-inserted for actual playback.

 

[KSTR]

And you can of course adjust the modulator noise slope so the it doesn't have that kind of sharp knee.

I was looking at a recorded signal where the modulator noise slope is fixed and given. I wanted to illustrate that the analysis tools used allow for much greater accuracy and bandwidth than was is seen in commercial DSD analyzers like MusicScope as it imposes its own DSD-to-PCM conversion and filtering for display and analysis, just as you said.

For DSD DAC implementation, I know ;-) that you know ways to reduce the insane RF content of the 1-bit stream, for example by using the conceptually simple trick of a sliding window average, creating a hardware rectangular FIR filter.

For the fun of it I looked at the effect of a sliding window average over 64 samples (in this ideal hardware simulation):

Blue: original DSD raw spectrum.
Cyan : 64-sample sliding window average applied to it, affording up to almost 40dB attenuation of the RF.
Red : Additional 50kHz 3rd-order analog lowpass filter on top of that which now has much less RF to deal with... which is of course not the only advantage of the FIR filter, as we know ;-)

 

[kongwee]

Exactly.
This is what a DSD stream, a single bit semi-randomly toggling at many MHz all the time (11.3MHz in this case), looks like:
View attachment 193005
The audio content is "buried" in that RF noise carrier, the noise is modulated with the audio content, a pulse density modulation (PDM). Because a bit is never repeated for more than two samples the vast majority of the energy is moved upward, near Nyquist frequency.

Direct editing on a 1-bit stream is limited to:
- cutting/trimming
- polarity inversion (flip bits)
- time inversion (bits backwards)

We always have to remember that a DSD stream is really just the captured modulated recording signal with the goal to keep the DAC hardware simple, in the most elemetary sense only an analog low-pass filter is needed to remove the carrier from the digital input signal. Editing was never within the scope of that total ADC-->DAC chain where the signal is tapped off after the ADC, stored (on file/SACD), given to the customer and the re-inserted for actual playback.

Still can't comprehend. DSD is just 0 and 1 digitally, just like PCM 0 and 1. PCM you can but DSD you can't. So you are saying you can't convert the graph you shown into actual waveform just like PCM for DAW work? DSD file is a captured audio file just like PCM. Both format have time stamp, sampling frequency, bit rate....etc are already fix in metadata. You can't display waveform in DSD in one second captured while PCM you can? How to explain you can transform between DSD and PCM with the same Red Book file with zero different digitally? Again DSD file is already captured data, you can't eq, put effect on already digitally and offline but you can on PCM? It doesn't make sense. For converter, I already sold as miska mentioned, DSD converter is just simple and better. DSD and PCM gonna give the same low noise audio frequency at much higher sampling rate.

 

[Miska]

I was looking at a recorded signal where the modulator noise slope is fixed and given.

Modulator you usually find used in such has usually gentler slope than the SoX one - if you used that one for the plot like it seems.

So there would be even less noise left.

 

[MAB]

But the thing is....it sounds fabulous (the surround version I mean...have barely ever listened to the stereo).

(and yes, I've looked at its surround waveforms too---it's loud)

Yes, it's loud! And I love it in all of it's oversaturated glory. And I listen in stereo, so we have all the bases covered here...

 

[Miska]

The audio content is "buried" in that RF noise carrier, the noise is modulated with the audio content, a pulse density modulation (PDM).

You should consider it more like PWM for practical cases. Since in hardware implementations adjacent bits of same value don't change state.

Because a bit is never repeated for more than two samples the vast majority of the energy is moved upward, near Nyquist frequency.

Well, this his incorrect, there are more than two adjacent samples of same value. It depends on your modulation index. With maximum allowed modulation (75%) you get +3.1 dB DSD level. They actually check this at SACD production, and will refuse to make you the disc if you don't stay within the specs.

Good to remember this if you are making DACs with ESS chips or other that have 0 dBFS PCM = 0 dB DSD. It means that with DSD inputs the DAC output can go 3.1 dB over the max you can get with PCM. For this reason, AKM has 0 dB DSD = -3.5 dBFS PCM.

Direct editing on a 1-bit stream is limited to:
- cutting/trimming
- polarity inversion (flip bits)
- time inversion (bits backwards)

You can also do for example delays, like I do for speaker distance processing in multichannel speaker setup when running in Direct SDM mode.

We always have to remember that a DSD stream is really just the captured modulated recording signal with the goal to keep the DAC hardware simple, in the most elemetary sense only an analog low-pass filter is needed to remove the carrier from the digital input signal. Editing was never within the scope of that total ADC-->DAC chain where the signal is tapped off after the ADC, stored (on file/SACD), given to the customer and the re-inserted for actual playback.

Since doing good modulation in a tiny DSP like DAC chip, it also allowed to move that rather complex and heavy process outside of the playback gear. And that still applies today.

There were three types of DAWs for editing DSD. Now only one left on market, if you don't count my tool that is more like complementary add-on tool to the process with Pyramix. Sure I wish I would have time to make full fledged DAW. (Pyramix allows you to do cutting/trimming/copy-paste etc bit-perfect on DSD)

 

[Miska]

And no, software doesn't keep working if OS revisions break it.

That's why Linux is so great. Because you have all the freedom to keep it working!