Octave Music Don Grusin High Resolution Music Analysis (Video)

[TCD333]

Thanks for your reply.
In response to the argument that OOB noise is not an issue, I will let others argue that issue with you. But, what are the reasons for using DSD for the end user? Amir has posited that there is no benefit to using DSD for the end user. The reason that this thread goes round and round in circles, in my opinion, is because no one has definitively refuted Amir’s conclusion. People post a lot of measurements and technical jargon, but none of it is clear evidence that would result in a different conclusion. If you could help me by posting a succinct reply explaining the reason to reject the conclusion that is the topic of this thread, I would greatly appreciate it.

IME people use / listen to DSD for subjective reasons, IOW they prefer the sound and I am referring to 128 / 256.

TCD

 

[pkane]

IME people use / listen to DSD for subjective reasons, IOW they prefer the sound and I am referring to 128 / 256.

This is where you can expect to be asked for evidence that DSD 128/256/512/1024/16384/65536 makes any audible difference.

 

[dualazmak]

You didn't read mine. As stated the OOB noise is attenuated through the DAC if it is designed correctly.

All of those plots are showing OOB noise of the music files, not what comes out of a DAC.

TCD

You did not read my posts cited there.

I actually measured the signal comes out of a DAC (OKTO DAC8PRO), and I confirmed that the UHF noises actually go into SP drivers, tweeters and super-tweeters. I decided, therefore, to use -48 dB LR high-cut (low-pass) filters at 25 kHz all the way in my upstream digital DSP (EKIO).

Generally speaking, DAC should be transparent for all the signals including UHF components to be fed to SP drivers, as far as no intentional high-cut (low-pass) filters included.

 

[TCD333]

You did not read my posts cited there.

I actually measured the signal comes out of a DAC (OKTO DAC8PRO), and I confirmed that the UHF noises actually go into SP drivers, tweeters and super-tweeters. I decided, therefore, to use -48 dB LR high-cut (low-pass) filters at 25 kHz all the way in my upstream digital DSP (EKIO).

Generally speaking, DAC should be transparent for all the signals including UHF components to be fed to SP drivers, as far as no intentional high-cut (low-pass) filters included.

My apologies Dualazmak, you had included a lot of links and I missed that Okto DAC was in the chain.

You have covered some interesting ground, a part of which may even point to the DAC's self generated OOB noise but it's hard to tell at a glance.
Do you have a link to wide band plots of that DAC playing various PCM formats vs DSD64/128/256?

Are you concerned about damaging that very nice T925A super tweeter? The power levels of even -30dB WB noise (integrated over BW)
would be extremely low (something like 0.002 watt or less) at very loud music levels.

Still - the OOB noise issue is, for me at least, quite an interesting point of discussion because (in part) it relates back to the actual DAC architecture, digital and analog filter design. As Miska was trying to point out, all modern DS DACs are basically some form of DSD, namely an array of resistors (or current sources) switching on / off at high frequency.

TCD

 

[TCD333]

This is where you can expect to be asked for evidence that DSD 128/256/512/1024/16384/65536 makes any audible difference.

DSD65536... haha DSD256 is max you want to go. Most DAC's I've seen data for seem to deteriorate after this.
WRT evidence, IME mastering folk will generally do blind testing, most home listeners wont. So it goes

If I remember correctly, Amir has visited Bruce Browns facility that does a ton of DSD mastering work and is a proponent of the format.

Amir, do you have any comments on that visit - do they conduct DBT?

TCD

 

[amirm]

WRT DSD OOB noise:

a/ DSD64 (2.8224MHz) does inherently have OOB noise, depends on the order of noise shaping used. Any properly designed AC's that play native DSD will have an FIR LPF which is inherent in the DAC OP architecture. Check the data sheets. After that follows an analog LPF of 3rd or 4th order.
What's left will be too low to harm any electronics or speaker.

I don't need to check any datasheet. I played the DSD content with two DACs: RME ADI-2 DAC and Topping, both of them spit out just about all of that noise. So did a very expensive Marantz SA-10 SACD player. As I have explained, it is political suicide for the DSD spec to mandate filtering just above audio band as it would then lose any definition of "high-res audio" in the minds of its buyers. So what you say is simply not true and is wishful thinking.

Here is Marantz again:

RME:

And topping:

 

[dualazmak]

Thank you @amirm for your continuing follow up.

Hello again @TCD333,

@amirm also kindly wrote here;
I have done this and garbage most definitely is output by the DAC. if it did not, it would not be a proper dac. Increasing sample rates needs to increase the bandwidth.
Having my measurement data, of course I am joining amirm on this common sense point .

I wrote such a high amount of UHF noises would be "possibly" harmful (and useless, meaningless) for our tweeters and super tweeters. I also pointed they would be highly possibly harmful for our pets on which you have not yet responded.

One another (important!?) point would be, because of my age-dependent slight decline in hearing ability over 8 kHz (even though I still have much better hearing capability over average score of my age group), the sound over 20 kHz (or over 15 kHz?) is almost meaningless at least for me; even though I have variable upward FR response tuning system for 6 kHz to 25 kHz compensating my slight age-dependent high Fq hearing decline.

 

[TCD333]

I don't need to check any datasheet. I played the DSD content with two DACs: RME ADI-2 DAC and Topping, both of them spit out just about all of that noise. So did a very expensive Marantz SA-10 SACD player. As I have explained, it is political suicide for the DSD spec to mandate filtering just above audio band as it would then lose any definition of "high-res audio" in the minds of its buyers. So what you say is simply not true and is wishful thinking.

Here is Marantz again:

RME:

And topping:

WRT political suicide: I think the idea was actually allow greater BW for filter transition band so as to avoid brick wall digital filters
and still attain good image attenuation. From memory the recommended filter for DSD(64) was at 50kHz.

I can't argue with your plots which clearly point to designers either a/ not adequately accounting for DSD64 content or b/ not considering that
amount of US noise a problem. It's hard to say.

It is worth mentioning that a Zero OS Multibit DAC at 44.1k with minimal or no OP (analog) filtering will have similar level of ultrasonic tones and
significant noise to boot. There would be many thousands in use globally.

TCD

 

[dc655321]

I think the idea was actually allow greater BW for filter transition band so as to avoid brick wall digital filters
and still attain good image attenuation.

Not with DSD.

It is worth mentioning that a Zero OS Multibit DAC at 44.1k with minimal or no OP (analog) filtering will have similar level of ultrasonic tones and
significant noise to boot. There would be many thousands in use globally.

Ah, an equally broken approach!

 

[Blumlein 88]

WRT political suicide: I think the idea was actually allow greater BW for filter transition band so as to avoid brick wall digital filters
and still attain good image attenuation. From memory the recommended filter for DSD(64) was at 50kHz.

I can't argue with your plots which clearly point to designers either a/ not adequately accounting for DSD64 content or b/ not considering that
amount of US noise a problem. It's hard to say.

It is worth mentioning that a Zero OS Multibit DAC at 44.1k with minimal or no OP (analog) filtering will have similar level of ultrasonic tones and
significant noise to boot. There would be many thousands in use globally.

TCD

The problem is nearly all DSD DACs and SACD players I've seen tested over the years have the same ultrasonic noise profile as what Amir is getting for DSD64. So no it is not atypical, rather typical. The exception is the devices which do something about it ( I don't recall seeing one that did). Is there an example of one that did have this filtered out?

 

[KSTR]

I've clearly shown in post #527 that even just a simple 3rd-order analog Butterworth filter at 50kHz suffices to reduce the RF noise of a heavily noise-shaped DSD256 stream. More than enough to filter the RF content down to levels of typical 20kHz content, and that was for a classical piano track with almost nothing at 20kHz to begin with.

Therefore, even with an agressive noise-shaped DSD64 that same filter alone still is good enough to afford the bulk of required reduction. Actual DACs have additional digital filters that reduce the noise even further, along the lines of what I've shown in post #555.

==> A complete non-issue. No tweeters will ever be killed, no downstream electronics will get upset.

 

[TCD333]

I've clearly shown in post #527 that even just a simple 3rd-order analog Butterworth filter at 50kHz suffices to reduce the RF noise of a heavily noise-shaped DSD256 stream. More than enough to filter the RF content down to levels of typical 20kHz content, and that was for a classical piano track with almost nothing at 20kHz to begin with.

Or 4th order Bessel / Linear Phase filter (Bessel slightly less steep than Butterworth but has better transient / phase response)

Therefore, even with an agressive noise-shaped DSD64 that same filter alone still is good enough to afford the bulk of required reduction. Actual DACs have additional digital filters that reduce the noise even further, along the lines of what I've shown in post #555.

==> A complete non-issue. No tweeters will ever be killed, no downstream electronics will get upset.

My point WRT Zero OS DACs - I'm not aware of anyone that has had issues with speakers or amps.

 

[amirm]

I've clearly shown in post #527 that even just a simple 3rd-order analog Butterworth filter at 50kHz suffices to reduce the RF noise of a heavily noise-shaped DSD256 stream.

The OP is about DSD64 which is also the largest library of download content in DSD. It thins out greatly from there on. I have a bunch of DSD64 content but just one or two DSD256. In sharp contrast, there is plenty of 96 kHz PCM content for download which has no ultrasonic noise stuffed right above 24 kHz.

 

[TCD333]

Not with DSD.

DSD was always promoted as having better transient response without ringing. This really is a similar thing.
A slower (longer passband) linear phase LP filter will have better transient response sans ringing.

BTW, Yes - I do understand Shannon / Nyquist namely that any material which would theoretically cause such (pre / post) ringing
is not encoded... theoretically.

Ah, an equally broken approach!

WRT Zero x OS, technically broken or not, a LOT of people use them and that doesn't appear to be changing.

TCD

 

[KSTR]

Topping D10 Balanced, using single ended output signal here, so 2Vrms @ 0dBFS.
Input signal is a 11kHz sine at -40dBFS in DSD64, generated with Audition and SoX-DSD, using the SDM-6 modulator (modulators range from SMD-4 to SDM-8, with SDM-8 having the most extreme noise shaping).

Output signal bottom trace is time waveform, we can see that the 11kHz signal is present but already overlaid with RF of about the same RMS magnitude.
Top trace is the FFT, showing corresponding peaking levels around 90kHz a bit below that the 11kHz payload signal level.

This tells us the RF energy at the output is ballpark about the same level than a continuous sine at -40dBFS (1/100).
A 100W@4R amp (=20Vrms output) will thus push about only 200mV (10mW) of noise even into a non-padded 4R-tweeter when whatever volume control is set to produce non-clipped 100W from a 0dBFS signal -- which means already an insanely loud volume setting.

==> non-issue.

 

[Blumlein 88]

DSD was always promoted as having better transient response without ringing. This really is a similar thing.
A slower (longer passband) linear phase LP filter will have better transient response sans ringing.

BTW, Yes - I do understand Shannon / Nyquist namely that any material which would theoretically cause such (pre / post) ringing
is not encoded... theoretically.



WRT Zero x OS, technically broken or not, a LOT of people use them and that doesn't appear to be changing.

TCD

I think the idea lots use non-oversampling DACs hard to support. There are some, but a tiny percentage of available DACs. No chance supporting it is not a broken design because it is.

 

[TCD333]

I think the idea lots use non-oversampling DACs hard to support. There are some, but a tiny percentage of available DACs. No chance supporting it is not a broken design because it is.

My clients range from musicians to Pro Engineers to Audiophiles to just audio enthusiasts.
Of all the audiophiles, more than 50% use non OS DACs. Looking at how many MB 0 x OS are available ATM, it's fair to say a lot are in use.
It's a broken design technically and I don't use one personally because of that.
However I have heard some systems running 0 x OS DACs that sounded incredible.
The technical part of my brain struggles to rationalize this but that's just the way it is.

TCD

 

[dualazmak]

In any way, I do not like to see the VU meters of DSP for high and super-high Fq stay at abnormal high level due to the inaudible (and possibly harmful) UHF noises, as shared in my post here;

Consequently, at least I myself always use -48 dB/Oct high-cut (low-pass) LR filters at 25 kHz in DSP fully eliminating these UHF noises which often included in DSD (and/or HiRes PCM) tracks.

Just for your notice, my music library of about 25,000 files consists of mixture of various formats;

16-bit/44.1kHz CD ripped aif,
24-bit/192kHz down-sampled or up-sampled aif,
24-bit/96kHz flac,
24-bit/192kHz flac,
1-bit/DSD64(1x) 2.8MHz dsf,
1-bit/DSD128(2x) 5.6 MHz dsf,
1-bit/DSD256(4x) 11.2 MHz dsf,

and now JRiver MC feeds all of the tracks in 88.2 kHz 24 bit (i.e. max. 44.1 kHz Fq window in 2-ch stereo) by on-the-fly conversion into EKIO for crossover/EQ processing. As I have high-cut (low-pass) filters at 25 kHz, max 44.1 kHz in L & R channels are more than enough.

These days, when I would download purchase music tracks, therefore, 24-bit/96 kHz or 24-bit/196 kHz should be quite OK (for my ears and brain, and also for my audio system), and usually they are much cheaper than DSD tracks.

 

[bennetng]

Octave Music Don Grusin High Resolution Music Analysis (Video)

https://audiosciencereview.com/forum/index.php?threads/octave-music-don-grusin-high-resolution-music-analysis-video.31620/post-1122924 Oh yeah, and no DSD involved Miska has posted measurements on a few forums over the past many years. Like he said, you need to look around. Even various...

www.audiosciencereview.com

It is amusing that everyone put the focus on ultrasonic noise without paying attention to stuff below 20kHz. Even Miska's trusted AKM DAC with his trusted modulator and his trusted Prism analyzer showed that DSD256 is about 3dB noisier than 44.1k PCM, so no more excuse about ADC aliasing and such. If the rise of <20kHz noise is really caused by ADC aliasing, then it simply means DSD256 is still noisier than 44.1k PCM, even in a "Miska approved" environment. Archimago got similar results as well:

Do you need linear power supply for DACs?

The tests were done with the same SoX DSD signal with the same modulator settings, so all differences were originated from the DAC. All RMAA tests were done with the ADI-2 ADC, so with this factor being constant, all differences were originated from the DACs under tests. You just don't seem...

www.audiosciencereview.com

Regardless of audibility, the whole DSD upsampling thing is simply a waste of time.

 

[TCD333]

View attachment 196694

It is amusing that everyone put the focus on ultrasonic noise without paying attention to stuff below 20kHz. Even Miska's trusted AKM DAC with his trusted modulator and his trusted Prism analyzer showed that DSD256 is about 3dB noisier than 44.1k PCM, so no more excuse about ADC aliasing and such. If the rise of <20kHz noise is really caused by ADC aliasing, then it simply means DSD256 is still noisier than 44.1k PCM, even in a "Miska approved" environment. Archimago got similar results as well:

Here is a PCM1792 (Mono mode) outputting -150dB signal in DSD64 mode. For DSD256 shift the noise lower and to the right. No in band problems to see here.