Why 24/192 is a bad idea?

[bdiament]

Hi Bruce,

We have found this as well. All converters have a "sweet spot" where they function better at one base sample rate (ie: 44.1 or 48). "Most" of the coverters I"ve used sound better at freq of 88.2/176.4/352.8

I attribute this to clocking within the unit.

One thing I haven't found yet is a converter that does 4x rates well but does not do 2x rates well. It seems the clocking (as well as the analog stage performance at wide bandwidth) seems to determine the sample rate "ceiling" for the device. The best ones I've found don't seem to "care" whether the rate is based on 44 or 48. Others seem to show more attention paid (or skill designing) in one area over another.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

[MarinJim]

Have you looked at the specs for the DAC chip itself? If it uses delta-sigma conversion (and maybe even if it doesn't), the DAC chip itself oversamples. Many manufacturers of entire DAC units use no oversampling filters prior to the DAC chip itself, and thus call their units NOS, but that's a bit misleading.

Yes, very good point.

 

[Orb]

Have you looked at the specs for the DAC chip itself? If it uses delta-sigma conversion (and maybe even if it doesn't), the DAC chip itself oversamples. Many manufacturers of entire DAC units use no oversampling filters prior to the DAC chip itself, and thus call their units NOS, but that's a bit misleading.

Also worth considering that there are others that bypass the DAC chip imbedded oversampling and use an external OEM oversampling chip, while others go on to develop their own bespoke utlising FPGA chips-processor.
All comes back to implementation and whether done very well,average, or poorly.
The imbedded oversampling function within a DAC chip does not guarantee quality though.

Cheers
Orb

 

[Orb]

And to add, one aspect that may have an implication when discussing various sampling rates (but I do not think mentioned yet) is that of pulse width distortion, which may have implications due to hardware performance-spec at higher sampler rates (not an issue at 44.1 but possibly when going higher).
This ties partly into what I mention with regards to receiver chip, which can also be embedded into DAC.
I think that was/is one of the embedded issues or challenges relating to the reference ESS Sabre DAC chip, but please not not quote me on that as could be wrong
Anyway, pulse width distortion is one of the low level hardware considerations that may also affect quality and definitely an implementation consideration when discussing 24/96 and higher.
Why IMO the discussion about 24/192 or 24/96 is too complex to be reduced to simple conclusions, interesting to discuss I agree...... if it is not sidetracked by "but is it audible and prove it is" debates

Cheers
Orb

 

[rbbert]

My comment about embedded oversampling in DAC chips related only to the issue of non-oversampling, not the inherent audio qualities of any chip.

There are of course several different ESS Sabre chips, and as mentioned different implementations possible for each chip. Generalizing about the sound of the "ESS Sabre" chip based on only one of those possible combinations would be inappropriate.

 

[amirm]

I guess it depends on the definition of "in the studio". The 3 kHz tone is definitely not there during the recording. However, it may be there in the monitors when doing the mix. That being said, unless the monitors have exactly the same non-linearity as your speakers (highly unlikely), what you're hearing at home is not what they heard from the studio monitors.

I addressed that already. In general, we do not hear exactly what is in the studio whether we are talking about in-band or ultrasonic tones. What is certain though is that if the distortion exists in the studio and you completely remove it before giving the material to me, for sure you have magnified the difference.

What makes you believe that your speakers have the same non-linearity as the studio monitors? After all, the power of the 3 kHz tone depends only on the speaker non-linearity.

Per above the biggest difference will come if you completely remove any of those tones in the version I get. Now, that may be a good thing. But it may also be a bad thing. For that reason, I keep making a point that you ignore in response after response: that I can perform the same signal processing at home, removing the ultrasonics. When *I* do it, I have a choice of leaving it there, removing a bit of it, or a lot of it. And even change these decisions on a track by track basis after performing a listening test. There is no way someone doing this upstream without access to my sound system can do the same.

Interestingly, I haven't seen any speaker manufacturer include "ultrasonic non-linearity" in its specs.

They don't give you that for in-band either. It is only through third-party testing such as Soundstage network that we get any distortion data. In that sense, the issue remains that we can't replicate what is in the studio 1:1.

So maybe you're getting 10 dB more than what's in the studio monitors, or maybe 10 dB less. Who knows. OTOH, if you record the output of the studio monitors (or more simply, simulate the same non-linearity), then you can at least have hear the same thing.

If I have a -6 db fs 1 Khz tone in the recording, I may be hearing it at -10 dbfs and in the mixing studio, -3. Your solution seems to be to remove the 1Khz tone because of this variation even though when the music was mixed, they would not do that. That is a strange argument per above in that if there is any harm, I can handle it myself.

Bottom line is the logical point I made: you have not shown any reason to truncate data in the studio. If you and others are personally concerned about these things and think the person mastering the track knows more than you do about your system, then by all means, get the 16/44.1 version of it. They almost always offer this version in parallel to the high-res version and if they didn't, you can perform the conversion yourself. If you think the bar needs to be even lower at compressed signals, they offer that also in the form of MP3 and AAC. To the extent someone is paying a premium for music, then we should give them all the bits.

Of note, the content owners have responded to consumer demand for much smaller files than CDs by lowering the quality bar. Folks here say let there be an offer above CD too. Even if it is all psychological, they still want that choice. I see no technical or business reasons to not do that.

Fortunately for all the protests the labels are providing these high-res files. So at the end of the day, your argument is not something that is resonating with the content providers or their consumers. It all becomes a forum tug-o-war with nothing constructive. Bob's article came out when folks were deciding the formats for DVD-A and SACD. Now it is all done and we have moved on.

The above point is important: somehow we managed to get the content owners to release their "master" bits. People in video would kill for the 4K versions for video but that is not happening. In music, there was a time that the labels would throw you out the door if you asked for their "masters" to be released, much less with no copy protection! But here we are with the right thing happening: customer wants it and the content owner appears fearless for a change and servicing them. So unless you work for the Inquirer and want to aim for publicity, we need to move on.

 

[amirm]

Also worth considering that there are others that bypass the DAC chip imbedded oversampling and use an external OEM oversampling chip, while others go on to develop their own bespoke utlising FPGA chips-processor.
All comes back to implementation and whether done very well,average, or poorly.
The imbedded oversampling function within a DAC chip does not guarantee quality.

I am glad you made this point as I wanted to mention this earlier . It was said that the filter in the DAC is a walk in the park now because there is "digital oversampling" in the DAC. Just because something is in the DAC and is "digital" it does not mean at all that it is perfect. Indeed, we can't perform perfect processing there. Interpolation has accuracy limits and implementation constraints. What the designer chooses is a trade off. He makes the filter design easier but then builds in distortion that was created by the interpolator.

Best version of this comes in the forum of DACs with asynchronous sample rate converters which advertise that they have done away with jitter. Yes, by resampling to a fixed target clock, they sharply reduce source induced jitter. But then they have a similar problem in that they need to constantly figure out the ratio of incoming clock to outgoing and that adaptiveness (sp?) creates its own forum of distortion. Unless you know how to measure that, you may be fooled by small jitter measurements. Same with upsampling in the DAC.

 

[arnyk]

And if interested, here is a paper by Julian Dunn which gives very solid research & findings for the possible reasons why a higher sampling rate than 44.1KHz should sound better.
"Anti-alias and anti-image filtering:
The benefits of 96kHz sampling rate formats for those who
cannot hear above 20kHz"

The cited paper is a technical summary and seems to shed no light on issues related to audibility.

This paper should satisfy those who deny that there are thresholds of audibility, and solemnly believe that everything sounds different.

There's no doubt that all sorts of measurable and theoretically predictable changes can happen to music on its way through audio components, but which of them is actually audible seems to be of the essence.

 

[Orb]

My comment about embedded oversampling in DAC chips related only to the issue of non-oversampling, not the inherent audio qualities of any chip.

There are of course several different ESS Sabre chips, and as mentioned different implementations possible for each chip. Generalizing about the sound of the "ESS Sabre" chip based on only one of those possible combinations would be inappropriate.

True but I raised it for same reason you did about oversampling DAC chips, could be argued that is inappropriate as it is generalising as there several utlising bypass of oversampling.....
In reality both yours and mine statements have merit and are not about generalising but importantly considerations, whether that involves embedded functionality used such as oversampling, or in my case expanding upon this and and providing some insight as to the hardware layer of challenges, which are seen and still around in modern day and even reference level DAC Chips, and other related architecture chips.

Edit:
I think we both need to consider where we may be coming from on this discussion, as 24/192 a good idea? topic covers such a broad scope.
For me it entails all aspects, including actual native hirez music files, communication, both up and downsampling, filters,etc and not just oversampling DAC chips but also their architecture implemented such as bypassing certain functions that also leads to consideration for embedded-external functions, in essence the hardware level and associated coding and performance-traits.
So that is why I feel there is merit in both our posts, and they are far from inappropriate (lumping yours with mine on that ).
Anyway, that post is also relevent because of pulse width distortion and/or tolerance thresholds that can be exceeded.
But each forum member can take what they want from what is posted.
Cheers
Orb

 

[Orb]

I am glad you made this point as I wanted to mention this earlier . It was said that the filter in the DAC is a walk in the park now because there is "digital oversampling" in the DAC. Just because something is in the DAC and is "digital" it does not mean at all that it is perfect. Indeed, we can't perform perfect processing there. Interpolation has accuracy limits and implementation constraints. What the designer chooses is a trade off. He makes the filter design easier but then builds in distortion that was created by the interpolator.

Best version of this comes in the forum of DACs with asynchronous sample rate converters which advertise that they have done away with jitter. Yes, by resampling to a fixed target clock, they sharply reduce source induced jitter. But then they have a similar problem in that they need to constantly figure out the ratio of incoming clock to outgoing and that adaptiveness (sp?) creates its own forum of distortion. Unless you know how to measure that, you may be fooled by small jitter measurements. Same with upsampling in the DAC.

Thanks Amir glad someone agrees
Your point on implementation constraints and choosing trade-offs is great as this touches exactly on some of the discussions John Westlake shared with the public on his Audiolab 8200 design, he specifically mentions from what I remember implementation constraints and setting priorities for implementation that affects performance or traits in some of his posts.
I should mention it is not just John W but a very talented colleague who is also involved with designing the various filters.
Cheers
Orb

 

[amirm]

The cited paper is a technical summary...

Summary of what Arny? It is a reprint of an AES conference paper: http://www.aes.org/e-lib/browse.cfm?elib=8446. It is the same copy so it is not summarized in that manner.

... and seems to shed no light on issues related to audibility.

There are references and discussions around that. But yes, it is not a listening test. Such a test would follow correct understanding of the design so that we can use the right stimulus in the listening tests -- something that unfortunately is almost universally missed in creation of double blind tests.

This paper should satisfy those who deny that there are thresholds of audibility, and solemnly believe that everything sounds different.

You mean people who go to AES conference fall in this category? Or is it that he is providing an explanation of how higher sampling rates -- independent of whether we hear ultrasonics -- can be beneficial if one looks at how a DAC works. We all can use more education of how our systems work.

There's no doubt that all sorts of measurable and theoretically predictable changes can happen to music on its way through audio components, but which of them is actually audible seems to be of the essence.

Lacking extensive listening tests we have a choice of shooting blind or looking at science and measurement. This paper is the latter. Readers are welcome to choose which camp they want to be in .

 

[Ethan Winer]

Why would a couple of FFTs prove that 'the only thing that changed is the noise floor' ?

LOL, because the noise floor is the only difference!

If you see anything else that's different, please point it out to me. But simply saying "you're wrong" with no further explanation is insufficient. I proved my point with science and hard data. If you can prove your point I'd love to be educated.

--Ethan

 

[jkeny]

Originally Posted by jkeny
And if interested, here is a paper by Julian Dunn which gives very solid research & findings for the possible reasons why a higher sampling rate than 44.1KHz should sound better.
"Anti-alias and anti-image filtering:
The benefits of 96kHz sampling rate formats for those who
cannot hear above 20kHz"
The cited paper is a technical summary and seems to shed no light on issues related to audibility.

This paper should satisfy those who deny that there are thresholds of audibility, and solemnly believe that everything sounds different.

There's no doubt that all sorts of measurable and theoretically predictable changes can happen to music on its way through audio components, but which of them is actually audible seems to be of the essence.

The cited paper is a technical summary and seems to shed no light on issues related to audibility.

This paper should satisfy those who deny that there are thresholds of audibility, and solemnly believe that everything sounds different.

There's no doubt that all sorts of measurable and theoretically predictable changes can happen to music on its way through audio components, but which of them is actually audible seems to be of the essence.

I'm not so sure you are correct as I believe you missed this bit in the paper which refers to audibility

The production of pre-echoes from filter ripple was reported by Lagadec and
Stockham [6],. They found the pre-echo due to a filter ripple of ± 0.2dB with a span
of 23Hz corresponded with echoes of -32dB at ±40ms - which was found to be quite
perceptible even with untrained listeners. A question is raised of how much echo,
and in particular how much of the un-masked pre-echo, can be permitted without
producing a degradation in the highest quality reproduction system?

So why not try pre-echo audibility yourself?

I can post links to two files I have prepared which will allow you test if you can hear, on your system, such audio artifacts as is described in that paper. In other words a pre-echo @-39dB added 20 samples in advance of the signal. I have run this test blind with a number of people now & most correctly identify the file with pre-echo (without peaking). Maybe others would like to take this test also & report their findings?

 

[NorthStar]

What are the specs of the dCS Ring DAC?

 

[opus111]

LOL, because the noise floor is the only difference!

But FFTs are histograms, there's no 'floor'. Added to that, an FFT is a time exposure.

If you see anything else that's different, please point it out to me. But simply saying "you're wrong" with no further explanation is insufficient.

In your opinion, for sure. Yet we're doing science are we not? In which case its the claimant who has to support, not the detractor.

I proved my point with science and hard data. If you can prove your point I'd love to be educated.

How about providing evidence for your claimed love? Like putting up a 'properly implemented' DAC in response to my earlier request? I'd love to educate you on that but I see no curiosity on your side.

 

[jmvalin]

I addressed that already. In general, we do not hear exactly what is in the studio whether we are talking about in-band or ultrasonic tones. What is certain though is that if the distortion exists in the studio and you completely remove it before giving the material to me, for sure you have magnified the difference.

I see your point. So I assume that when you shop for loudspeakers you always check that the distortion measure is at least 1%. Otherwise, you might miss out on the the non-ultrasonic IMD that was also present in the studio monitors. Avoid like the plague any speakers that have lower distortion than standard studio monitors.

 

[MarinJim]

Does anyone besides me love a good dog fight?

 

[NorthStar]

Mouse fight here, in Canada; we luv those. :b ...We always buy a lot of traps, with our groceries.

 

[amirm]

I see your point.

I think you do too! But this being a public forum and us both being males means you will never admit it . That if I get high-res content, I can get rid of the ultrasonic or not. The choice will be mine based on listening tests. You taking it out brings no value to me and forces my hand one way.

So I assume that when you shop for loudspeakers you always check that the distortion measure is at least 1%. Otherwise, you might miss out on the the non-ultrasonic IMD that was also present in the studio monitors. Avoid like the plague any speakers that have lower distortion than standard studio monitors.

For me to "miss out" on something means that the person mixing the content did indeed hear ultrasonics. And that he had a positive experience as a result of it. In that regard, we better all group together to quantify and measure it so that we can reproduce it at home. Would you like to join us in that? Or should we conclude that it is best to not go there?

 

[xiphmont]

I am compelled to think you have not read his paper after all.

Bob addressed it much later (if I didn't read the paper, why did I agree with the comment you made about sampling frequency being tradable for greater bitdepth?). But that's not what you were talking about in your post.

Also the larger question of whether the graph is even valid or not is unanswered. It's comparing integrated spectral energy of an unknown variable window to pure tone ATH amplitude. That's not kosher without more information than is presented, nor am I aware of any research confirming direct correlation between narrowband noise and pure tone ATH. This is a plausible concept on its face, one reason I hadn't thought to question it, but that's not the same as it being true. It is definitely true that I can make those TPDF levels move up or down at will depending on the window I choose.

Also where are those ripples in the TPDF line coming from? There shouldn't be any.

So, I think I'm going to write him and ask if he can offer a clarification. I suspect he'll be happy to answer.

Also: Adobe Audition manual: not authoritative on any subject

But if the capture is done at 24 bits which is usually the case when music is recorded/mixed these days, you would be creating distortion products due to decimation to 16 bits. If you captured and stayed in 16 bit mode then yes. But such is not the case.

The whole point of dither is to eliminate distortion when downquantizing (and the extension of the usable dynamic range at the bottom end can be considered in the same light, otherwise when the signal level dropped to > .5 LSB the harmonic distortion becomes infinite). A rigorous description/proof (and a good read with examples):

http://uwspace.uwaterloo.ca/bitstream/10012/3867/1/thesis.pdf