Another major look at MQA by another pro.

[Fokus]

2 minutes ago, PeterV said:

Nobody over here knows how the MQA encoding and decoding process really works.

 

We know a lot, including the fundamental principles at the base of every aspect of this trickery.

 

As Mansr mentioned before: this contrary to you, who does not seem to know/understand a lot of this.

 

[Fokus]

28 minutes ago, Shadders said:

What we do know, is ...

 

Your knowledge is also still lacking...

 

Apart from the DRM, A(uthentication), and the let's-obsolete-all-and-revive-the-market aspects MQA is a scheme that passes a 2x/4x/8x recording through a 1x channel, without using orthodox, steep AA and AI filters. MQA claims steep filters are evil, and so it invites alias/image distortion plus increased noise in a trade to remove the filters.

[Fokus]

6 minutes ago, PeterV said:

This was for me the ultimate listening test, since if even this DAC is improving it's soundquality, it is clear that the format is not a hoax or marketing scam.

 

You are aware that at least some MQA files have been derived from totally different masters, perhaps even different mixes, than the standard releases that are out there?

 

You are aware that undecoded MQA contains the hidden signal crud that contaminates the base signal, potentially in an audible way?

 

6 minutes ago, PeterV said:

 expressing myself to them that a Time Domain measurement would be of interest for all of us. Would it be possible to achieve this? Well, if MQA would do such measurements themselves, no one will believe it I suppose. So it is an interesting task to find a technique where the audible time-smear reduction of MQA can be verified with objective measurements in Air

 

What do you need measurements for? You can derive a decent guess of the overall impulse response of an MQA chain quite easily with the information that is now publicly available. Assuming that you start from a 8x recording (DXD), concatenate the IRs of the ADC and two of the leaky filters Mansr/Archimago have published. That's about it.

 

You are aware that looking at IRs of systems with a bandwidth > 20kHz and imagining that a nice Dirac impulse is the holy grail has not much to do with actual music, nor with how the human auditory system works?

 

Let's do a thought experiment. Take a 44.1k ADC and DAC. Remove the anti-aliasing and the anti-imaging filters. Put them back to back.  Feed in a nice, perfect impulse. Observe the output and swoon. Or drool.

 

Now feed it some music, and listen ...

 

6 minutes ago, PeterV said:

Nowadays much DSP software is available like Trinnov, Dirac, DEQX and others which are manipulating loudspeaker performance in the Time & Frequency domain heavily..

 

Yes. And ironically, if MQA have their way all these systems are obsolete, and MQA have to be paid dues (and royalties) in order for these systems to function again.

 

But yes, frequency/time correction is done to speaker systems. With great effect. But this is smack in the middle of the audible band, with sub-systems with non-linear frequency response and with significant inter-driver delays.

 

6 minutes ago, PeterV said:

 I am only interested in the end result and why my ears ( and others) seem to like it..

 

No. You are only interested in whatever confirms your notion that MQA is better, and that this is because of something you call time smear (and which you do not understand, at all).

 

6 minutes ago, PeterV said:

there should at least be one of them which contains a relatively isolated single tone.. Just use that and do A/B comparison.. or is this an impossible route?

 

You want people to listen to one note and conclude from this that MQA is better? Or do you want people to look at the trace of a single note???? If the latter: you won't see ringing or blurring. Really.

 

6 minutes ago, PeterV said:

.do you realise that during the introduction of the CD we also have been 'seized' by Sony and Philips..?  What is your problem with a sophisticated, but proprietary technology which is improving the sound.

 

CD was entirely new, building on nothing that came before. As such it was simply necessary for customers to buy into it totally.

MQA is a flavour of PCM, behind locked doors, threatening to obsolete open PCM and all of its tools (i.e. software and hardware, see my remark on DSP above). And this for no use. Even if it really sounded better, it would be of no use.

 

6 minutes ago, PeterV said:

Just take tome and listen to it first..I am convinced you will understand why I am so positive and have the urge to express myself especially to this and other pre-occupied and very negative persons and groups.

 

Oh, I did. I do have an MQA DAC. I do have listened to it. In many cases the MQA was just as good/bad as the original. In some cases the MQA was better, but clearly from a different master, so the gains would pass equally well through a non-MQA chain.

 

6 minutes ago, PeterV said:

 the best way to understand really what MQA is all about.

 

I know perfectly well what MQA is about.

 

[Fokus]

17 minutes ago, soxr said:

I am here because we found an open source way to handle MQA files

 

What?

 

You take undecoded MQA and upsample it with a MP filter. That's not 'handling MQA files'.

 

[Fokus]

23 minutes ago, soxr said:

 

Yes it is.

 

 

 

How does it then differ from handling non-MQA files?

 

You are just trying to direct attention to your own products.

How some people feel undecoded-MP-upsampled MQA sounds when compared to the (alleged) DXD master is not very relevant, and is not relevant at all when discussing MQA, because your 'method' does not employ anything that is specific to MQA.

 

[Fokus]

33 minutes ago, soxr said:

The OSS world is not amused with MQA.

 

And rightfully so.

 

 

33 minutes ago, soxr said:

The Sox method uses similar minimum phase upsammpling as the MQA renderer, but does not apply the weird anti-ringing filters.

 

How can you use 'similar minimum phase upsampling' as the renderer, while at the same time not applying the 'weird anti-ringing filters'? The MP upsampling filters and the WARFs are the same in the renderer. Moreover, the renderer only acts on a core-decoded 2x input (ignoring for the moment the pathological case of 1x MQA), where HF signal levels are lowish and thus the output of the 'weird' filters is not that weird at all, if you follow the reasoning behind MQA.

 

You ignore the 2x to 1x folding stage. If you replay MQA with your method, starting from the 1x file, you throw away half of the signal bandwidth. How can this be a valid method for 'decoding' MQA?

 

[Fokus]

24 minutes ago, mansr said:

If you're not paid by MQA, you are the epitome of a fanboy. If you are paid, you're a shill.

 

How about a techo-challenged overenthusiastic audiophile, the kind the magazines and dealers must love to bits, because they can sell anything to them?

[Fokus]

Just now, mansr said:

Those "pathological" 1x files make up about half the MQA tracks on Tidal.

Yes. But they are not the reason MQA exists, and nor are they related to the fundamental trickery behind it.

 

It is as if the markering dept saw Craven/Stuart's ideas and then asked "how can we broaden the umbrella to include, and cash in on, 1x as well?"

 

[Fokus]

1 hour ago, soxr said:

Please provide proof that these ultrasonics can be heard and do matter.

 

Allow me to be confused. You start this with a comparison to DXD, and then suddenly ultrasonics do not matter anymore.

 

1 hour ago, soxr said:

Even MQA mentions this:

 

They claim subjective transparency without actually conveying ultrasonic information (above 2x rate, below 2X they do convey it), but with the proviso that there has been no steep filtering at the 1x band edge.

 

Then you come along, take an undecoded MQA file, thus at 1x, and replay with a steep  filter and claim a similar result.

 

Oh well.

 

Ever considered this: if you play an undecoded MQA file through your Adagio it behaves as a NOS DAC (well, it is a NOS DAC) and you get the usual treble droop and imaging (and potential IMD).

If you play the MQA file through the Brooklyn and defeat MQA, yet retain the filter named 'MQA' (which is the MP filter with rather poor stop band suppression, see Stereophile measurements) you get almost as much imaging.

If you upsample with sox -vsM you get no imaging, no IMD...

 

 

[Fokus]

28 minutes ago, PeterV said:

 suddenly someone is making use of the MQA encoded files, upsamples them with his own quasi MQA up-sampling algorithm and it is as good as DXD...!  It actually proves that MQA encoding is upgrading ...

 

What soxr is doing has nothing to do with MQA and with the data encoded in an MQA file, despite his claims.

 

[Fokus]

7 minutes ago, PeterV said:

  It is possible though,

 

 

Of course it is possible. Within the MQA ecosystem.

 

This means that DSP manufacturers need to take MQA licenses and pay per-unit royalties.

 

This means that existing DSP solutions, out in the field, have now three options:

-the manufacturer takes a license and provides a firmware upgrade (free or paid)

-no upgrade feasible, keep using it, replaying undecoded (i.e. compromised) MQA files

-no upgrade feasible, dump it all in the landfill, run out to buy a new shiny 3-letter-word toy

 

I am sure that for a significant amount of DSP systems out there an upgrade will never materialise.

 

[Fokus]

23 minutes ago, PeterV said:

We really need to understand the way our ears and brain are being 'stressed' by unnatural sound effects such as the pre- and post- ringing actually is..

 

Sigh.

 

The only 'evidence' for the audibility of filter ringing at ultrasonic frequencies is in marketing blurbs.

 

Real listening tests tell a different story:

https://www.stereophile.com/features/106ringing/index.html

 

http://www.pinkfishmedia.net/forum/showpost.php?p=2089923&postcount=32

 

[Fokus]

3 minutes ago, PeterV said:

, but the cumulation of all filter errors is..   as mentioned in this article http://www.theabsolutesound.com/articles/beyond-high-resolution/?page=3

 

I quote

 

" As Fig. 3 shows, temporal blur is cumulative; each successive stage can inadvertently spread transient energy over a wider and wider interval. It’s not uncommon for a signal to have been subjected to a cascade of eight filters by the time it’s gone from microphone to loudspeaker; the damage is significant, and yet each individual stage isn’t that bad in isolation. "

 

This is predominantly about analogue bandwidth limitations in the chain. The worst offenders, by far, are microphones, analogue tape machines, and the loudspeakers. MQA is doing nothing for these. Showing this graph in the context of MQA is mere spreading of FUD.

 

Oh wait, didn't I mention marketing blurbs in my previous post?

 

A competent digital production should only see two, perhaps three instances of steep, ringy filtering:

1) ADC anti-aliasing

2) post-mastering donwsampling (in case of a lower-rate distribution format)

3) DAC anti-imaging

 

The listening tests I referred to cover this flow.

 

Anything with more steep filtering than the above is either a flawed production, or the result of aesthetic decisions. If it is a production fault then MQA can't cure it (because it can only attack a single filter signature, if at all). If it is an aesthetic decision then MQA should not touch it.

 

[Fokus]

41 minutes ago, soxr said:

It's also insinuated in one of the sponsored-by-Bob papers that the best way to do it is at the end of the chain.

 

True. But only in the general case when the only weapon at hand is an AI or AA filter and one decided to stick to the rules.

 

MQA has more weapons: MQA opts for downsampling from 8x or 4x to 2x with barely an anti-alias filter, thus violating the sampling theorem. Presumably their AA filter is just enough to kill the ringing of the recording ADC (for an 8x original at 192kHz, a frequency so high that not much in real music will trigger it) and to keep the blatant aliasing caused by this lack of filtering below a pre-set threshold in the audible band (say -80dB).

 

This procedure is the key to MQA-speak 'deblurring'. I don't think that simply by inspecting an undecoded MQA file, or its core-decoded 2x equivalent, one can confirm or deny that this procedure is actually executed on commercially available music.

 

[Fokus]

27 minutes ago, soxr said:

x2 does NOT change the steepness of the original waveform (compared to DXD source)
x4 or x8 with their weird anti-ringing does, but it's very subtle

 

We don't care much about the steepness. MQA's promise is to get rid of, in their view, evil ringing.

 

To validate this you need to identify a recording that contains this ringing in the original master. I think you have the best chance with looking at 96kHz masters. Once you find this ringing (I wish you good luck!), see how core-decoded MQA compares.

 

You are not interested in looking at waveforms of x4 or 8x renders. There is so much imaging that these are very misleading. Plus, they don't add any information over the 2x version.

 

[Fokus]

54 minutes ago, PeterSt said:

 Generally there are three options :

1. In the source file as we can obtain it from e.g. Tidal;

2. During the first "unfold" when 44.1 or 48KHz is transferred to 88.2 or 96KHz and which can happen in PC software;

3. During the rendering stage which can only happen in MQA certified hardware.

 

 

Why still all this confusion???

 

Since end of 2014 the available MQA documentation has clearly stated what their aim is, and how it can be achieved. The aim is to pass x4..x16 originals through an x2 channel (ignoring for the moment the folding to x1), without using steep filters in the chain (filters which would cause 'blur'). How this is achieved is simply by downsampling from x4..x16 to x2 with a suitably-chosen very leaky AA filter. The filter selection is ideally done with an eye on the total amount of aliasing this generates below 20kHz, attempting to keep it below a certain threshold.

 

What is not clear is

a) does MQA today really do as above? (But why not, as it is simple ... Occam etc.)

b) and if so, are the AA filters really selected according to the signal at hand, or is it one-size-fits-all?

 

a) and b) are nearly impossible to verify without access test signals.

 

Also less clear is what MQA does to material that is originally at x2. Since there is no downsampling step they cannot use the non-blurring leaky Anti Alias filters. They could possibly apply an apodising low-pass filter to the x2 signal, be it LP or MP. Such additional filtering should be visible when spectrally comparing a core-decoded x2 MQA file to its x2 original. (This comparison is hindered by MQA not always using the same masters as are commercially available in hi-res.)

Alternatively MQA could apply an all-pass filter with excess group delay positioned at the original ADC's frequency of ringing (MQA have a patent application for this). This would move the pre-ringing to a post-impulse position (at least in the marketing graphs). Since the spectrum is unaffected this is not detectable with spectral comparison.

 

For x1 originals MQA could apply many tricks that correct for the ADC's sins, but none of these can be named deblurring since the bandwidth limitation is too severe. But let's forget about x1.

 

Back to x2 and higher: Remember that MQA wants to avoid steep filters anywhere in the signal path. So after unfolding to x2 the system must upsample with leaky Anti Imaging filters. We know that this is being done.

 

 

So where is the deblurring being done?

During the conversion from the hihi-res master to the MQA file, using leaky anti aliasing.

How is the deblurring preserved after core-decoding to x2?

By replacing the DAC chip's own anti imaging filters with leaky MQA anti imaging filters.

 

 

 

 

[Fokus]

Essentially, yes.

 

The method I describe, which is taken from MQA, gives an end-to-end impulse response of ~50 us for a 384kHz master, whereas an orthodox downsampling to 96kHz and back up to 384kHz would give 500 us or so.

 

(Of course what MQA don't tell is that with all that aliasing and imaging going on our textbook impulses are temporarily dislocated to the tune of 2.6 us, whereas normal resampling is accurate down to ns.

 

If not this, then what is the deblurring, in your eyes?

 

 

[Fokus]

1 hour ago, Jud said:

 

That sounds rather interesting (though I don’t know if this “temporary dislocation” results in anything we can hear), and I wish I understood it.  But I have a feeling it would require a few courses I didn’t take in college.

 

You know that the temporal accuracy of a correctly sampled system is very high. Once you allow aliasing this no longer holds. MQA allows aliasing up to a certain level. Thus ...

 

 

[Fokus]

12 hours ago, soxr said:

 In this paper they recommended using only ONE apodizing filter, at the end of the playback chain.

 

So I still assume (3) until otherwise is proven. My waveform analysis does not show otherwise.

 

 

MQA is not about apodising, and the render filters are too weak and at the wrong transition frequency to be apodising.

[Fokus]

11 hours ago, crenca said:

 

Fokus, could this measurement be recovered in the total energy of some (relatively broad) time period (thinking of what Charles was saying upstream)?

 

If the aliasing is allowed to affect the baseband (<20kHz), then the timing in the baseband is damaged, and no subsequent low-pass filtering ('broad time period') will recover it. That's what my intuition says.

 

Anyone with a good audio editor and programmable sample rate conversion software can run an experiment.

 

1) make a silent stereo file at x16 and insert 1 unit impulse in each channel, with a relative displacement of 1 sample. This is the analogue-world signal (*)

 

2) downsample to x8 with a traditional half-band linear phase filter. This is the DXD master.

 

3) downsample to x2 with a very weak filter such as the MQA render filters shown earlier here. These filters have a very very wide transition band, and only little attenuation at Nyquist. This is the MQA master before folding to x1.

 

4) ignore folding to x1/unfolding to x2 (because this step is ideally lossless, one problem at a time!)

 

5) upsample to x8 with a very weak filter such as the MQA render filters etc etc.

 

6) observe how the two impulses have been distorted: they no longer look the same, because an aliasing system is not time-invariant, and the 1-x16-sample temporal dislocation of the input signal has been corrupted.

 

The distortions observed depend on how much aliasing is allowed to creep in in step 3). With normal music and MQA rules this can be presumed to be low. But even so, the mechanism is inherently time-distorting. But it does show nice narrow end-to-end impulse responses for the marketing leaflets and the audio magazines.

 

A bit perverse?

 

For fun you could add a few extra steps. You could approach the hearing limit of an adult with a minimum phase low pass at 14kHz. Then you could model the cochlea as a filter bank of increasing passband widths, with the final filters 2kHz wide or so.

 

 

 

(* And due to aliasing this unit impulse is invalid. In a more elaborate experiment one could start at x32 and add some minimum phase pre-filtering to the impulse, taking the response as reference analogue-world impulse for the rest of the test.)

 

[Fokus]

Yes, applying non-linear processing in a closed signal space (i.e. a sampled system) causes immediate aliasing. This has been known for ages, but many early plug-ins were not aware of it, hence the band-aid of upsampling-processing-downsampling.

 

But we are talking MQA music delivery. What processing do you mean with " So it makes sense to upsample once more (hence the renderer part), and process that, because you have more sample points to work with, and the errors get distributed over more samples. "? There is no further processing. Either you still don't understand MQA, or you choose to express yourself in an entirely alternative language.

 

Like PeterSt is doing all the time ...

 

We really can miss this extra layer of confusion.

 

[Fokus]

16 minutes ago, PeterSt said:

 

Not quite, because the filtering is the upsampling process at the same time.

 

Indeed.

 

Even more exactly: the 'weird' filters are the anti-imaging filters for the upsampling process.

 

(And in the original embodiment of MQA these filters were meant to be loaded into the DAC chip, substituting the DAC chip's standard filters. This then would limit MQA to the handful of chips that support this. I can imagine that in some present or future incarnations the upsampling with these filters is actually done in the USB receiver processor preceding the DAC chip, if it is of adequate power.)

[Fokus]

6 hours ago, mansr said:

Do the words "Eastern District of Texas" mean anything to you? 

 

Yep. Been on the receiving end, once. Amazing how much hidden meaning there is in a simple word like 'transistor', and how that meaning can be turned into financial gains...

[Fokus]

How often do we have to repeat that MQA contains one stage of folding. All else is leaky downsampling and upsampling. MQA themselves won't deny this.

[Fokus]

On 8/12/2017 at 11:18 AM, PeterV said:

I just would like to understand how these arguments can be achieved..  How much time-smear does a DXD recording contain..? and when such a recording is played on a DXD capable DAC, how much is the Time-Smear in air ?  These are sinply fundamental questions to understand the importance of the presence or absence of Temporal Blur artefacts in music!

 

Start studying the sampling theorem, including its formal proof. Don't stop until you really get it.  Then look into DSP, and into AA and AI filters in particular. This done, have a look at the impulse responses of popular ADC and DAC chips. Ask yourself how relevant these are, considering the legal status of the stimulus. Look into the spectral distribution of typical music. Study the human auditory system. How does the basilar membrane react to a steep wavefront. How wide is an ERB in the upper octave of the ear response. What is the temporal implication of this?

 

once you have done this you can come here for any further questions. Until then COMMENT EDITED FOR PERSONAL ATTACK. - CC