MQA creator Bob Stuart answers questions.

[Blumlein 88]

Fascinating WRT high frequency signals. You do realise that DSD format produces lots of ultrasonic noise?

From 2L recordings http://www.2l.no/hires/
Joseph Haydn: String Quartet In D, Op. 76, No. 5 - Finale - Presto
Engegård Quartet

DSD64

View attachment 28134

same track at DSD128

View attachment 28135

2L with Morton Lindberg make some excellent recordings. They record in DXD. They are interesting too because I don't find much ultrasonic content in their music. There is far more ultrasonic content in the DSD versions, but as you show it is typical DSD noise.

They favor DPA microphones. While DPA does make a couple models rated to 40 khz, those aren't the ones 2L prefers. I'm sure the 20 khz rated DPA's respond somewhat to more than 20 khz, but likely have a declining response. 2L has used some ribbon pairs to highlight soloists in a few of their recordings. Those are only flat to 5 khz, down 5 db at 10 khz and rolling off around a 1st order slope above that.

Comparing the most transient features in that Haydn at 192 vs 44, nothing in it strains the 44. Maybe they use the microphones to in effect implement a slow roll off filter so no ringing can possible occur. Of course that goes against the super transient momentum transfer idea vs there is some noise up there and some very low level harmonics to the music. Much too low in level to move the human eardrum enough to be perceptibly relevant.

 

[mansr]

AFAIK upsampling is done by adding zero samples and then lowpass filtering the result. This is described in Richard G Lyons Understanding Digital Signal Processing, a very good book. According to him intrapolating is not a good idea.

Inserting zeros and applying a low-pass filter is interpolation. A crude filter with triangular impulse response produces linear interpolation. With a sinc filter you get proper band-limited interpolation.

 

[bennetng]

2L with Morton Lindberg make some excellent recordings. They record in DXD. They are interesting too because I don't find much ultrasonic content in their music. There is far more ultrasonic content in the DSD versions, but as you show it is typical DSD noise.

They favor DPA microphones. While DPA does make a couple models rated to 40 khz, those aren't the ones 2L prefers. I'm sure the 20 khz rated DPA's respond somewhat to more than 20 khz, but likely have a declining response. 2L has used some ribbon pairs to highlight soloists in a few of their recordings. Those are only flat to 5 khz, down 5 db at 10 khz and rolling off around a 1st order slope above that.

Comparing the most transient features in that Haydn at 192 vs 44, nothing in it strains the 44. Maybe they use the microphones to in effect implement a slow roll off filter so no ringing can possible occur. Of course that goes against the super transient momentum transfer idea vs there is some noise up there and some very low level harmonics to the music. Much too low in level to move the human eardrum enough to be perceptibly relevant.

As far as I know this track from 2L should justify a 96kHz distribution format in terms of frequency content. At least I can play it at half speed to hear what is going on:
https://www.audiosciencereview.com/...es-dsd-sound-better-than-pcm.5700/post-128714

Beyond that, it is in the realm of synthesizers.

 

[Blumlein 88]

Here is the DXD stereo of that 2L Haydn quartet. I've set the spectrogram to go to light gray at -70 db FS. So anything below -70 db won't be shown. As you can see precious little past 25 khz. Even 20-25 khz is -60 db. So played back with peaks of 120 db, there is nothing at high enough level for the most extended human listener to hear as their thresholds at and above 20 khz are over 100 db SPL. These in this recording would fall 40 db short or worse.

Same settings for 44.1/16 bit

 

[March Audio]

Here is the DXD stereo of that 2L Haydn quartet. I've set the spectrogram to go to light gray at -70 db FS. So anything below -70 db won't be shown. As you can see precious little past 25 khz. Even 20-25 khz is -60 db. So played back with peaks of 120 db, there is nothing at high enough level for the most extended human listener to hear as their thresholds at and above 20 khz are over 100 db SPL. These in this recording would fall 40 db short or worse.

View attachment 28138

Indeed. Sergei seems to confuse/conflate the potential of transient shock waves at high levels (nuclear bomb haha) with the real world levels found in music. Its a fascinating (read bizarre) link he has made there.

Has anybody got any 192kHz tracks that are PCM so we can look at HF without the noise of DSD? I have a number of 96kHz but not 192.

 

[Krunok]

 

[March Audio]

found some chesky tracks.

By 46kHz its down to -90 dB

this one is hardly any better

nor this. its all very predictable.

 

[bennetng]

View attachment 28141

 

[Blumlein 88]

Indeed. Sergei seems to confuse/conflate the potential of transient shock waves at high levels (nuclear bomb haha) with the real world levels found in music. Its a fascinating link he has made there.

Has anybody got any 192kHz tracks that are PCM so we can look at HF without the noise of DSD? I have a number of 96kHz but not 192.

I do. They are on a stored hard drive. But I recorded them with medium condenser mikes that have some response beyond 20 khz, but aren't really wide band microphones. They do have some very heavy transient sound on them at very high level. But they don't have much above 15 khz at any high level. The transients were from some wooden blocks or sticks being struck. These are very loud. The based fundamental is around 7 khz, and some harmonics above it. I didn't have any wide band microphones at the time.

I also have some music that had high level chimes recorded with some pencil omnis. I'll see if I can find those, but they may be on that same stored hard drive.

 

[Blumlein 88]

found some chesky tracks.

By 46kHz its down to -90 dB

View attachment 28143

The spectrograms on Music Scope aren't those for the entire track? Reason I ask is that could be deceptive. You can have short lived transients that represent so little of a 3 or 4 minute recording they don't show up much in aggregate. When I show those in Audacity I pick an FFT width for the spectrogram that is equivalent to roughly 50 milliseconds as that is something like how the ear might respond to them.

 

[Krunok]

Has anybody got any 192kHz tracks that are PCM so we can look at HF without the noise of DSD? I have a number of 96kHz but not 192.

I have Yazoo - Best of (2018).

Here is how it looks (entire track scan):

 

[March Audio]

The spectrograms on Music Scope aren't those for the entire track? Reason I ask is that could be deceptive. You can have short lived transients that represent so little of a 3 or 4 minute recording they don't show up much in aggregate. When I show those in Audacity I pick an FFT width for the spectrogram that is equivalent to roughly 50 milliseconds as that is something like how the ear might respond to them.

Not of the entire track, but true I dont know the FFT time. I can try them in audacity, but I wouldnt expect to see much different. As we both know, the mics just arent high enough bandwidth to record it in the first place.

 

[Blumlein 88]

Everything above -70 dbFS

Everything over -120 db FS.

 

[March Audio]

The spectrograms on Music Scope aren't those for the entire track? Reason I ask is that could be deceptive. You can have short lived transients that represent so little of a 3 or 4 minute recording they don't show up much in aggregate. When I show those in Audacity I pick an FFT width for the spectrogram that is equivalent to roughly 50 milliseconds as that is something like how the ear might respond to them.

Adobe Audition short segment. Not much different

 

[Blumlein 88]

We need software for ultrasonic transient momentum transfer grams. If Sergei can supply us with one example maybe it will be clear. Well other than that nuclear bomb blast example.

 

[Krunok]

Everything above -70 dbFS
View attachment 28150

Everything over -120 db FS.
View attachment 28151

That looks like noise.

 

[March Audio]

We need software for ultrasonic transient momentum transfer grams. If Sergei can supply us with one example maybe it will be clear. Well other than that nuclear bomb blast example.

Actually the best way to do this is not with FFT due to the "integration" time. Just high pass filter the time waveform above a certain frequency and see what is left in the time domain.

 

[bennetng]

We need software for ultrasonic transient momentum transfer grams. If Sergei can supply us with one example maybe it will be clear. Well other than that nuclear bomb blast example.

https://hydrogenaud.io/index.php/topic,108864.msg948272.html#msg948272
https://hydrogenaud.io/index.php/topic,108864.msg948350.html#msg948350
Looks much better than MQA, no gap.

 

[March Audio]

OK a video of high pass filtering above 40kHz. Anyone hear anything?

Basically after filtering what was left of some cymbal hits (those two spikes) didnt get above -57dB. Good luck if you think you are going to hear, momentumise or perceive that in any way @Sergei .

Any speakers response (lack of) will reduce its level further.

 

[LTig]

Inserting zeros and applying a low-pass filter is interpolation. A crude filter with triangular impulse response produces linear interpolation. With a sinc filter you get proper band-limited interpolation.

You are absolutely right. I reread the chapter of Ricks book and have to admit that I cited wrong . What Rick mentioned as bad practice is to repeat the original sample instead of using zero values.