Octave Music Don Grusin High Resolution Music Analysis (Video)

[PeteL]

I have that album and it was recorded and produced in DSD 64, according to the booklet. Not sure why someone would pay extra for a DSD256 version of a DSD 64 master.
It's a very well recorded album that sounds good. Good recording of the orchestra and it has a sense of the venue in the recording. Nothing wrong with it.
In any case, most DACs that will playback a DSD file without converting it to PCM have hi-frequency filters built in when converting to analog. The noise shouldn't be an issue. If they convert to PCM they are also probably filtering the ultrasonic noise when the conversion to analog happens.

I don't know about most DACs, but according to post 192, at least 2 popular DSD capable dacs don't filter out the ultra sonic noise.

 

[firedog]

OK, ran some real-time analysis on the analog output of a couple of DACs using the same Don Grusin track ("So Nan Desu"). Analog output is captured while Roon natively streams DSD (verified in both DACs to be the case):

First RME ADI-2 DAC Version 2. Per above, it is setup to play native DSD which disables just about every functionality in the DAC. It has two filter settings. I tested both:

View attachment 191986

Measurement bandwidth is 250 kHz which is the reason there is a roll off on the right. The filters themselves don't seem to be doing much other than creating a notch at their bandpass filter settings.

I then grabbed a Topping DX3 Pro+. I can't find any non-PCM filters in it so ran with it as is. I also boosted the measurement bandwidth to 1 MHz so we see the full spectrum, unfiltered by AP input:

View attachment 191987

The ultrasonic noise is there in all its glory. Musicscope had actually truncated the spectrum at 88 kHz. So the real story is worse than what I showed in the video.

Really, there is no scenario I can see where all of that noise is removed as doing so means limiting bandwidth to 25 kHz or so. No way any DSD advocate wants to admit that is the real encoding range. They rather ship noise than to fess up to that!

AFAIR, your Matrix DAC automatically filters DSD. I think at 47khz or 50khz, one of them is the default.

 

[don'ttrustauthority]

Can we leave comments our of this? I bought a bunch of the Chesky recordings, and some of those have this characteristic you refer to where it seems like it's being made to show off our systems (not that there's anything wrong with that!). But let's be nice to the musicians. Frankly I would think they would want to know about this kind of thing since the money being charged for the recording seems very expensive. I forget how much Chesky charges, I usually buy lo res, but I doubt it was that much because if it was more than $20 I wouldn't buy it ever. I support the musicans with lo res.

 

[dasdoing]

PS Audio has a private music label called "Octave Music." I got a request yesterday to look into their high-resolution offering and see what they are doing in that regard. So I downloaded the Don Grusin "Out of thin air" album ($29) and performed analysis both on the DSD version and PCM. There is also some technical explanation of the formats and explanation for what we see:

you might wanna make an EQ cut around/above 400Hz when recording in this room, since there is a strong resonance there. not saying it is important but it will sound more "professional"

 

[Music1969]

OK, ran some real-time analysis on the analog output of a couple of DACs using the same Don Grusin track ("So Nan Desu"). Analog output is captured while Roon natively streams DSD (verified in both DACs to be the case):

First RME ADI-2 DAC Version 2. Per above, it is setup to play native DSD which disables just about every functionality in the DAC. It has two filter settings. I tested both:

View attachment 191986

Measurement bandwidth is 250 kHz which is the reason there is a roll off on the right. The filters themselves don't seem to be doing much other than creating a notch at their bandpass filter settings.

I then grabbed a Topping DX3 Pro+. I can't find any non-PCM filters in it so ran with it as is. I also boosted the measurement bandwidth to 1 MHz so we see the full spectrum, unfiltered by AP input:

View attachment 191987

The ultrasonic noise is there in all its glory. Musicscope had actually truncated the spectrum at 88 kHz. So the real story is worse than what I showed in the video.

Really, there is no scenario I can see where all of that noise is removed as doing so means limiting bandwidth to 25 kHz or so. No way any DSD advocate wants to admit that is the real encoding range. They rather ship noise than to fess up to that!

@amirm is that with ADI-2 in 'DSD Direct' mode enabled?

And why does the ADI-2 look much 'choppier' in the audio band?

Why it doesn't look same as the Topping up to 20kHz?

 

[MAB]

When 'audiophile' recordings first hit the street, or at least when I first recall them, you had MFSL half speed masters, and direct to disk represented by Sheffield, Umbrella, Crystal Clear and some others. Crystal Clear not only used direct to disk, but cut at 45 rpm. Often you had to buy them in hi-fi salons, as major retailers didn't stock them, for whatever reason.

Back then, there was a real need for it, because a lot of the product coming from the major labels was very poor in the QC department. DGG, RCA, and CBS had world class artists, but bottom of the barrel pressings. CBS had the moniker, 'cost before sound'.

With the exception of Mobile Fidelity (who used established artist's tapes), most DD records featured second-rate performances. So, you mostly bought the record for the sonics. To show off your system. Slumming the boutique salons, you heard Amanda McBroom... all the time. Unless they were playing that drum record. And with that, Sheffield couldn't even afford to hire Hal Blaine!

However it was, something like this--average artist with questionable recording technique? Doesn't seem to make much sense.

Agreed, except there were some very good artists (Tower of Power direct to lacquer recording by Sheffield as an example, IMHO). I think that the mistake is extending the audiophool methodology of listening to recording. The biggest sins in high end audio seem to be misapplication of technology, which is sort of what this thread is about (there is nothing inherently wrong with DSD, it is the application here that is messed up!!!) Same with recording. Sheffield Direct to Disk is a totally appropriate application of the technology when applied to Tower of Power (great band live, 125dB horn section to challenge the recording engineering team, and they have great music). That drum record, not so much...

 

[PeteL]

@amirm is that with ADI-2 in 'DSD Direct' mode enabled?

And why does the ADI-2 look much 'choppier' in the audio band?

Why it doesn't look same as the Topping up to 20kHz?

I believe they are just random parts of the song not the same exact spot. The scale is totally different as well. I do think it wold be nice to have it presented as a frequency response plot with an actual dsd test sweep, but I think the purpose there was to show the filtering of the Ultrasonics.

 

[Miska]

I don't know about most DACs, but according to post 192, at least 2 popular DSD capable dacs don't filter out the ultra sonic noise.

Looks like a regular DSD64 recording. And that's why you'd want to have a DSD256 recording. Or upsample DSD64 to DSD256 or DSD512 for playback. DACs work much better with DSD256 or DSD512 inputs. At DSD64 the noise is too close to the audio band.

For example with ADI-2 Pro ADC, the noise floor slope looks same for both 705.6/32 PCM and DSD256. Because it is essentially two 1-bit ADCs running in parallel at DSD256 rate.

 

[Miska]

I believe they are just random parts of the song not the same exact spot. The scale is totally different as well. I do think it wold be nice to have it presented as a frequency response plot with an actual dsd test sweep, but I think the purpose there was to show the filtering of the Ultrasonics.

Here is my first gen ADI-2 Pro, sweep 0 - 22.05 kHz, peak hold spectrum.

DSD256, DSD filter set to "50 kHz" (re DSD64, thus becomes 150 kHz at DSD256):

DSD256, DSD filter set to "150 kHz" (re DSD64, thus becomes 450 kHz):

And for comparison, same but with 44.1/32 PCM input, digital filter set to "sharp":

On this one you can see the images left by the 8x digital filter.

If you input 705.6k upsampled data, you get images further away and lower level:

Now the correlated image is down to similar peak level as you get with the 150 kHz DSD filter setting. While the 50 kHz DSD filter setting still gives lower level. And it is just decorrelated noise compared to the 100% correlated distortion with PCM.

 

[Miska]

And for comparison, Holo Audio Spring DAC, same as above, but DSD512:

 

[PeteL]

Here is my first gen ADI-2 Pro, sweep 0 - 22.05 kHz, peak hold spectrum.

DSD256, DSD filter set to "50 kHz" (re DSD64, thus becomes 150 kHz at DSD256):
View attachment 192173

DSD256, DSD filter set to "150 kHz" (re DSD64, thus becomes 450 kHz):
View attachment 192174

And for comparison, same but with 44.1/32 PCM input, digital filter set to "sharp":

View attachment 192177
On this one you can see the images left by the 8x digital filter.

If you input 705.6k upsampled data, you get images further away and lower level:
View attachment 192180
Now the correlated image is down to similar peak level as you get with the 150 kHz DSD filter setting. While the 50 kHz DSD filter setting still gives lower level. And it is just decorrelated noise compared to the 100% correlated distortion with PCM.

Informative, thanks.

 

[dadregga]

So, in the end, what REALLY matters, is how good the original recording captures the sound, and how well mastered it is....

Nothing had really changed in decades.

Yep. You buy masterings, not formats.

Everything above 44.1/16 is theoretically better, but not audibly so.

Like the "better sound" of vinyl, the "better sound" of high res recordings comes from the (likely) extra care taken during mastering, and the fact that there (tend to be) fewer downsampling "steps" taken between digital capture and the high res file itself, and has absolutely nothing to do with the technical capabilites of the format itself.

I prefer to own physical copies of the highest-res versions of good masterings that I can find - not because I can hear ultrasonics but because of the above.

If Blu-ray licensing wasn't significantly more expensive we could get PCM multichannel and high res pressed on those - and we could dispense with SACD and DSD.

Sadly SACDs are still cheaper to press so if you want to ship someone either multichannel or higher-than-CD res on a disc that can be played back by a transport, SACD is your only real option unless you want to pay extra for Blu-ray.

 

[Miska]

When people look at spectrum of DSD data, they tend to forget how PCM looks like at same bandwidth.

This is how the same 0 - 22.05 kHz sweep in RedBook PCM looks like, unfiltered, at same 88.2 kHz bandwdith used for original video:

You will need heavy digital filters to make it proper. Trying to convert it to analog at 44.1 kHz rate, with just D/A and analog filters is pretty much futile exercise.

DSD64 is not as futile, but difficult still. You can see this if you look at the early delta-sigma DACs that were running at the same 2.8 MHz rate using 1-bit. But with converter designed for this. Here's same signal, pure output of CS4328 DAC chip without additional analog filters:

You can see that there's no noise shaping bump visible. But naturally some analog filtering is needed to remove the sampling rate noise and it's harmonics.

Given that maximum input resolution to the chip is RedBook, you can see it still provides full 16-bit distortion free range in audio band:

There you can see slight hump at about 76 kHz left of noise-shaper.

Description from the datasheet:

What is not true is the "no need for further filtering". You will certainly want to have some nice 3rd order analog low-pass on the output.

 

[pinpoint_oxford]

View attachment 192095
None of this makes sense to me. How is DSD more digital than PCM, yet more analog?

More digital in being strictly binary in terms of its signal, yet more analog due to ease of decoding to analog.

 

[Blumlein 88]

When people look at spectrum of DSD data, they tend to forget how PCM looks like at same bandwidth.

This is how the same 0 - 22.05 kHz sweep in RedBook PCM looks like, unfiltered, at same 88.2 kHz bandwdith used for original video:
View attachment 192187

You will need heavy digital filters to make it proper. Trying to convert it to analog at 44.1 kHz rate, with just D/A and analog filters is pretty much futile exercise.

Unfiltered PCM is basically broken, so I don't see the relevance of that graph.

 

[MakeMineVinyl]

More digital in being strictly binary in terms of its signal, yet more analog due to ease of decoding to analog.

PCM is strictly binary also.

 

[pinpoint_oxford]

PCM is strictly binary also.

I meant that in that DSD being 1-bit its data store is binary whereas PCM has 16-24 bits of information. Obviously PCM is binary in the sense that everything on a computer is, but it has words that make up its signal.

 

[Miska]

Unfiltered PCM is basically broken, so I don't see the relevance of that graph.

So is unfiltered DSD. Although DSD is of course intended to be filtered with just analog filters. Main benefits are in time domain performance, because you don't need to battle with the time-frequency relationship problems related to decimation anti-alias filters.

Just wanted to remind that looking at Nyquist-band plots of digital domain PCM may make it easy to forget that there's actually stuff beyond Nyquist. Which matters greatly when you want to go back to analog. And because people tend to look at non-Nyquist sampling systems like DSD in a different way.

 

[slmccaul]

Just took a look at some of the DSD64 albums I got (not really many(... Lol every single one of them exhibit this issue. The fewer DSD128 surprisingly are a mixed bag (have it but at lower levels, and it's a bit weird). And one DSD256 is literally as if someone mastered/recorded each track differently. Some tracks have no issues, while another can have a massive hump (-40dB of constant noise and also it's very wide on the frequency spectrum).

So I just converted all these troublesome tracks to 24-bit 44.1 or 48, with a low pass filter at 22.05 kHz

Saved a ton of memory as well. One DSD256 10-minute recording clocked in at 1.6GB, when converted to 44.1 + 24 bit, it clocks in at 99MB

Goodness, and I thought I only had to worry about fake "high res" recordings. Don't know why I didn't bother checking any of these DSD/SACD's Ive had about.

Hello, would you mind sharing what application you used to do this, especially the filtering part? I would like to look into doing this for two reasons. Saving storage space and not sending HF noise to my components and speakers. Thanks for the great info.

 

[Astoneroad]

Did you try using ONLY P.S. Audio power cords on each piece of your equipt.? Perhaps in order to get the "most" out of an Octave recording, you need the right cables, the fatter and more outrageously expensive, the better chance your brain will like the music more. I'm sure all that noise that you read would have been filtered by any $5k power cord. If not, I got lots of them for sale, genuine aliexpress P.S. Audio cords, originally $4995, now $49.95... step right up... make it an even $100 and I'll toss in the latest DSD download.... hurry... hurry... hurry.

(Reposted from YT)