Archimago's MQA listening test results

[svart-hvitt]

I've already offered some thoughts on Archimago's thread over at audio-asylum. Not about the subjective audibility of transient blurring, but on what constitues a fair test of MQA. For example, it actually is not a fair test to utilize identically mastered tracks for a comparison with High-Res PCM. Identically mastered tracks would serve more to obscure any advantages of MQA rather than to reveal them.

The MQA files of 2L, subject of Archimago's test, have been carefully made by Morten Lindberg('s team), seemingly in collaboration with MQA and Stuart: http://www.2l.no/hires/documentation/2L-MQA_Comparisons.pdf

So this is probably as good as it gets for MQA in real-life.

The great misunderstanding of Lindberg (2L sell MQA files at a premium price so it may not be a "misunderstanding" at all), Stereophile ("the press") and many audiophiles is that they believe MQA is better than the original, which in information theory terms is an impossibility. Euphonically, everything can happen. I believe many people favor lo-res MP3 over CD and hi-res.

MQA is for consumers in many ways like an IQ test in audio, where your reasoning determines in which basket you end.

 

[Fitzcaraldo215]

I've already offered some thoughts on Archimago's thread over at audio-asylum. Not about the subjective audibility of transient blurring, but on what constitues a fair test of MQA. For example, it actually is not a fair test to utilize identically mastered tracks for a comparison with High-Res PCM. Identically mastered tracks would serve more to obscure any advantages of MQA rather than to reveal them.

Interesting, Ken. But, I am not sure I follow why. I will look for your comments on Archimago's site.

 

[Ken Newton]

The MQA files of 2L, subject of Archimago's test, have been carefully made by Morten Lindberg('s team), seemingly in collaboration with MQA and Stuart...

I've not seen details of exactly how those test files were engineered. We can assume, but that's not proper. The primary technical issue is how the anti-alias filtering was implemented for the two sets of test tracks. For example, if both sets of tracks were initially mastered utilizing brickwall anti-alias filtering, and then the MQA targeted set was remastered from the originally brickwall mastered set, then the MQA set is not time-domain optimized and would likely sound very similar the High-Res PCM targeted tracks.

 

[Ken Newton]

Interesting, Ken. But, I am not sure I follow why. I will look for your comments on Archimago's site.

The comment link below more thoroughly addresses my concerns about what constitutes an fair set of MQA versus High-Res comparison listening test tracks.

http://www.whatsbestforum.com/showt...il-DAC-and-MQA&p=470116&viewfull=1#post470116

 

[Fitzcaraldo215]

Agree, there is no apparent value in MQA for the end user. End user ends (sic!) up paying something for nothing.

Let’s get the rationale of MQA straight too:

MQA is majority owned by Muse Holdings and Reinet Investments. They have the same owner, and Muse owns Meridian.

Reinet is listed in Luxembourg with twin-listing in South-Africa. Reinet is owned and controlled by the Rupert family, one of the richest in South-Africa.

The biggest investment of Reinet, which makes out well over 3/4 of Reinet’s market value, is British American Tobacco. Today, Johann Rupert runs Reinet. His father, Anton, started the tobacco business. Big tobacco has a long history of deception («nicotine is not addictive» etc.) and their money making has had big negative external costs for many individuals and the society at large.

Rupert aim to convert and transform their tobacco money by divesting into new ventures. MQA is one such attempt of making tobacco money shine in new clothing. A look at Reinet’s share price indicates that the attempt to get out of tobacco has been little successful till now.

It’s up to every thinking man to judge if the Rupert family has a strong record of ethics, socially responsible investing and if MQA is a clear break with big tobacco’s track record of deception and setting their own money interests only before all else.

By the way, the Rupert family is also well known for their philanthropy. Maybe MQA is philanthropy.

Interesting, however I do not ever ascribe outright guilt by association alone. There may be much more to the story and to MQA's motives.

 

[Fitzcaraldo215]

The comment link below more thoroughly addresses my concerns about what constitutes an fair set of MQA versus High-Res comparison listening test tracks.

http://www.whatsbestforum.com/showt...il-DAC-and-MQA&p=470116&viewfull=1#post470116

Thanks, Ken.

 

[Jinjuku]

I've already offered some thoughts on Archimago's thread over at audio-asylum. Not about the subjective audibility of transient blurring, but on what constitues a fair test of MQA. For example, it actually is not a fair test to utilize identically mastered tracks for a comparison with High-Res PCM. Identically mastered tracks would serve more to obscure any advantages of MQA rather than to reveal them.

So you want the same mix track mastered in two different ways to make it fair? That's a real head scratcher.

I'll make sure to master same tracks differently. I'll also make sure the louder ones are only played back on the more expensive cable, and as the cost of the cable goes up, the track will get louder.

Where have I seen that before...

 

[Jinjuku]

I've not seen details of exactly how those test files were engineered. We can assume, but that's not proper. The primary technical issue is how the anti-alias filtering was implemented for the two sets of test tracks. For example, if both sets of tracks were initially mastered utilizing brickwall anti-alias filtering, and then the MQA targeted set was remastered from the originally brickwall mastered set, then the MQA set is not time-domain optimized and would likely sound very similar the High-Res PCM targeted tracks.

That's an issue to take up with Bob and their friends at 2L. It's not on Archimago to sort out any communication / best practices issues with Meridian and 2L. Archimago is playing with the cards dealt.

That's perfectly fair.

 

[Ken Newton]

So you want the same mix track mastered in two different ways to make it fair?

Exactly. MQA is an time-domain optimized format - whereas High-Res is typically an frequency-domain optimized format. Minimizing transient blurring is the primary performance objective of MQA. Any test which does not correctly reflect that is not properly revealing. What remains identical between the two versions is the original analog signal.

Therefore, two different sets of test tracks are needed, the MQA targeted version optimized for the time-domain, utilizing only the MQA anti-alias filter (no brickwall filters utilized anywhere before or after). The High-Res PCM or RBCD targeted version optimized for the frequency-domain, utilizing only an brickwall anti-alias filter. It's all about the anti-alias filter implementation. Additionally, MQA should be played back through an MQA enabled DAC to to take full advantage of it's end-to-end system optimization of the time-domain.

I'll make sure to master same tracks differently. I'll also make sure the louder ones are only played back on the more expensive cable, and as the cost of the cable goes up, the track will get louder

I know this is counter-intuitive. Much of DSP science is.

 

[Ken Newton]

That's an issue to take up with Bob and their friends at 2L. It's not on Archimago to sort out any communication / best practices issues with Meridian and 2L. Archimago is playing with the cards dealt.

Not blaming Archimago. Just wondering aloud whether those conducting listening tests of MQA understand what novel benefit it is intended to bring to the listening experience, and how to test for that novel benefit. Which is the minimization of transient blurring otherwise caused by use of brickwall anti-alias and anti-image filters.

 

[Jinjuku]

Exactly. MQA is an time-domain optimized format - whereas High-Res is typically an frequency-domain optimized format. Minimizing transient blurring is the primary performance objective of MQA. Any test which does not correctly reflect that is not properly revealing.

Therefore, two different sets of test tracks are needed, the MQA targeted version optimized for the time-domain, utilizing only the MQA anti-alias filter (no brickwall filters utilized anywhere before or after). The High-Res PCM or RBCD targeted version optimized for the frequency-domain, utilizing only an brickwall anti-alias filter. It's all about the anti-alias filter implementation. Additionally, MQA should be played back through an MQA enabled DAC to to take full advantage of it's end-to-end system optimization of the time-domain.

I know this is counter-intuitive. Much of DSP science is.

MQA isn't a mastering technique. It's a convolution technique to already existing files. If we follow your logic all the album's that have been MQA'd in the Time-Warner catalog are then of dubious value.

 

[Ken Newton]

MQA isn't a mastering technique. It's a convolution technique to already existing files. If we follow your logic all the album's that have been MQA'd in the Time-Warner catalog are then of dubious value.

Fixing already existing files is not the optimum implementation of MQA. That feature is only intended to help improve recordings which have already been engineered for the frequency-domain. Meaning, engineered utilizing brickwall anti-alias filters so to deliver maximum flat bandwidth.

Native MQA is, instead, time-domain optimized from mastering through to playback. Meaning, is a transient optimized system, from anti-alias through to the anti-image filter implementations.

 

[Don Hills]

Exactly. MQA is an time-domain optimized format - whereas High-Res is typically an frequency-domain optimized format. Minimizing transient blurring is the primary performance objective of MQA. Any test which does not correctly reflect that is not properly revealing. What remains identical between the two versions is the original analog signal.

Therefore, two different sets of test tracks are needed, the MQA targeted version optimized for the time-domain, utilizing only the MQA anti-alias filter (no brickwall filters utilized anywhere before or after). The High-Res PCM or RBCD targeted version optimized for the frequency-domain, utilizing only an brickwall anti-alias filter. It's all about the anti-alias filter implementation. Additionally, MQA should be played back through an MQA enabled DAC to to take full advantage of it's end-to-end system optimization of the time-domain.

I know this is counter-intuitive. Much of DSP science is.

Ken,
You appear to be saying that an MQA track has to be MQA end to end, from the initial A to D of the microphone signal to the D to A at the listener. This would exclude every piece of recorded music to date from MQA processing. And MQA themselves do not claim this. They claim that it is sufficient to process an already digitised track provided the characteristics of the ADC are known. They also describe how this processing is done with an apodising style filter (they don't call it that, but the patent is clear). There may be additional processing done, but there's no sign of it in their descriptions or patents.

On the DAC side, there is nothing more than an apodising (and leaky) generic set of filters, the filter for each track being apparently chosen by analysis at the original processing stage. The filter sets in the MQA DACs analysed so far are identical - no customisation for a specific DAC implementation. So it's quite valid to use a non-MQA DAC to play back Archimago's test files. His test took MQA encoded 24/192 files and "unfolded" them usng the MQA first unfold and then MQA filters. This produced 24/192 files to play back through a standard DAC. I hope you don't believe the final reconstruction filters in typical 24/192 DACs somehow prevent the full benefits of MQA from shining through. A few minutes spent measuring their time domain behaviour should correct you if you do.

 

[Ken Newton]

Ken,
You appear to be saying that an MQA track has to be MQA end to end, from the initial A to D of the microphone signal to the D to A at the listener. This would exclude every piece of recorded music to date from MQA processing. And MQA themselves do not claim this. They claim that it is sufficient to process an already digitised track provided the characteristics of the ADC are known. They also describe how this processing is done with an apodising style filter (they don't call it that, but the patent is clear). There may be additional processing done, but there's no sign of it in their descriptions or patents.

Yes. I suspect that MQA have initialy focused their attention on the 'correction' feature because of the established catalog this opens up to being beneficially impacted by MQA, increasing the probability of general market success of the format. Bob Stuart's AES writings on MQA, however, make clear the optimum implementation is with newly recorded music, not with remastering of existing catalog.

On the DAC side, there is nothing more than an apodising (and leaky) generic set of filters, the filter for each track being apparently chosen by analysis at the original processing stage. The filter sets in the MQA DACs analysed so far are identical - no customisation for a specific DAC implementation. So it's quite valid to use a non-MQA DAC to play back Archimago's test files.

No, not if that DAC isn't optimized for the time-domain. A DAC featuring a typical brickwall anti-image filter will blur transient signals, just as Meridian claims, thereby weakening the transient response performance objective of MQA.

I hope you don't believe the final reconstruction filters in typical 24/192 DACs somehow prevent the full benefits of MQA from shining through. A few minutes spent measuring their time domain behaviour should correct you if you do.

Perhaps, I'm missing what you are saying, but any brickwall digital reconstruction filter features extensive ringing. It has to in order to provide a brickwall frequency response as they are the Fourier transform of each other. I have actually measured the time-domain behavior of such DAC's. I've even designed and built a few myself. Increasing the sample rate does shorten the duration of filter ringing, but significantly ring they still do. The number of cycles of ringing is the same as when processing a lower sample rate, it's just that the frequency of ringing is higher, thereby giving a shorter ringing duration. Optimum MQA implementation is designed to smear transients to a lesser degree than does High-Res PCM utilizing brickwall filters.

 

[Jinjuku]

Fixing already existing files is not the optimum implementation of MQA. That feature is only intended to help improve recordings which have already been engineered for the frequency-domain. Meaning, engineered utilizing brickwall anti-alias filters so to deliver maximum flat bandwidth.

Native MQA is, instead, time-domain optimized from mastering through to playback. Meaning, is a transient optimized system, from anti-alias through to the anti-image filter implementations.

That's not MQA's marketing message though.

 

[Ken Newton]

That's not MQA's marketing message though.

They have a muti-pronged marketing message, depending on the target audience, because MQA has so many aspects to it. This can be marketing blessing or curse for them.

 

[Blumlein 88]

I somewhat agree with Ken Newton's criticisms, assuming I have not misunderstood them.

The best as can be determined the processing done in the test matches what happens with MQA DACs. But it actually isn't MQA vs the original. That is one problem.

Another is if MQA is said to be a smaller audibly identical version of the true high rez material (DXD for instance), then a 50:50 split would indicate success for MQA. That is only one of MQA's claims.

A separate claim, somewhat soft-pedaled recently, is the idea MQA can compensate for and fix time domain blurring of material already recorded. People don't seem to have found any evidence that is actually being done. That still doesn't mean it isn't possible even if not yet implemented. So that test would be something like 2L's RDAT based material vs MQA versions. The MQA process says it can unravel the problems with the gear used initially and result in an improved sound. In that case MQA getting a 50:50 result is a failure.

The idea MQA could retro-actively improve in terms of accuracy recordings is highly, highly dubious for 99.999% of all music available due the various levels of processing done along the way.

No one who puzzles thru MQA's claims needs more than 30 seconds to decide what a convincing demonstration would be. Even MQA says DXD has low enough time blurring not to be an issue. So step one present me with some DXD recordings and the MQA version that unfolds into 96 khz which is audibly equivalent. Let me hear those decoded by MQA gear to hear it is so. Then present me with the same recording in properly downsampled or parallel recorded form with brick-walled 44 or 48 khz rates to hear the degraded sound vs the other two due to the blurring. That is it, that would be convincing, that would be simple and if the product works as claimed that would be the only way I would have demo'd MQA the first couple years. Instead MQA has never done that demo, will not do that demo, done everything to obfuscate the situation and never allow that demo. Initially you only had MQA vs MP3 which is insulting. These guys have spent so much time dancing around that issue it points to a con job in my book. It might not be so, but those people are not that stupid.

And that is of course one more problem with the Archimago test, we didn't get the CD versions. If MQA, and higher rez and CD all sound the same, well we all know what that means. So in the Archimago test 50:50 could be considered a success for MQA. As done MQA wouldn't claim improved sound in this situation it would claim audibly equivalent to higher rez. If CD had been included and none of the three sound different it would be a failure for the efficacy of both high rez and MQA or any other format claiming to equal high rez if high rez doesn't actually sound different than conventional normal sample rates.

 

[Don Hills]

... No, not if that DAC isn't optimized for the time-domain. A DAC featuring a typical brickwall anti-image filter will blur transient signals, just as Meridian claims, thereby weakening the transient response performance objective of MQA.
...
The number of cycles of ringing is the same as when processing a lower sample rate, it's just that the frequency of ringing is higher, thereby giving a shorter ringing duration. Optimum MQA implementation is designed to smear transients to a lesser degree than does High-Res PCM utilizing brickwall filters.

My point was, what is the ringing duration of a typical brick-wall 192KHz anti-alias filter? And how much ringing does a real-world musical signal cause at 192 KHz? Answering my own question: Not a lot, which means the filter doesn't have to be very good - something that MQA rely on. They fold (alias) what high frequencies there are down into the audible range so that they can later "unfold" those frequencies back to their original place in the spectrum. Which of course they do, but thay also "pollute" the HF spectrum with images of the audible frequencies. This, of course, is not mentioned. It's all designed to appeal to audiophile sensibilities.

 

[amirm]

I thought archimago emulated ma filter in his PCM samples. On the phone and can't check.

 

[March Audio]

No, not if that DAC isn't optimized for the time-domain. A DAC featuring a typical brickwall anti-image filter will blur transient signals, just as Meridian claims, thereby weakening the transient response performance objective of MQA.

Perhaps, I'm missing what you are saying, but any brickwall digital reconstruction filter features extensive ringing. It has to in order to provide a brickwall frequency response as they are the Fourier transform of each other. I have actually measured the time-domain behavior of such DAC's. I've even designed and built a few myself. Increasing the sample rate does shorten the duration of filter ringing, but significantly ring they still do. The number of cycles of ringing is the same as when processing a lower sample rate, it's just that the frequency of ringing is higher, thereby giving a shorter ringing duration. Optimum MQA implementation is designed to smear transients to a lesser degree than does High-Res PCM utilizing brickwall filters.

If this time domain issue is a concern to you then your emphasis is in the wrong place. Have you ever taken a look at what horrors a typical hifi speaker does in the time domain?