MQA Deep Dive - I published music on tidal to test MQA

[Oukkidoukki]

Okay, so I have been testing Tidal vs. Gobuz here couple of days.....comparison is very easy cause apps are in the same device........presentation is somehow thinner in Tidal, I dont know if it is the dynamics or what.......bass is more refined and stronger in gobuz, better sense of space also......sound wraps around your head in headphones, you dive deeper.......hihats are more real.........gobuz is closer to the sound of cd (which might funilly still win, havent tested).......tidals app is so far nicer , really like how it suggest artist and songs.....gobuz does this too, but so far not as well for my opinion..........by the way sound quality might get even higher if you download songs to the device, then u are not streaming over the net in real time, not sure about that yet.........I can hear difference with bluetooth headphones also, its not night and day ofcourse, but it is there...........So tidal interface might be tad better and faster ( probably because files are packed, which the can be heard) but it is not enough a reason for me to go back for now. Let’s keep testing.

 

[AdamG]

I like Tidal and will probably remain a Customer. I like the user interface and the new Tidal Connect feature. On top of that I take advantage of the generous Military 40% discount. I am old enough to realize I can’t hear the difference between the normal and higher quality tracks. At almost half price, it’s a keeper. MQA or not.

Discount link: https://support.tidal.com/hc/en-us/articles/360002657817-Military-Discount

 

[RichB]

My understanding of MQA explanations. As you are obviously much more knowledgeable in the subject than myself, please 'filter' )) what may be wrong:
They are not primary talking of the digital filters you mentioned (although they do analyze in depth the effects of decimation filters in the ADC process, a subject that's beyond my understanding), but instead of the analog ones with the known phase shift of harmonics they produce because of their capacitances and resistances.

In a very simplified explanation (sorry, not lecturing about Nyquist-Shannon, just trying to make myself clear): while you are converting an analog source to digital, there is a need to cut frequencies of the incoming analog signal before reaching the Nyquist frequency (NF), 22.05 Khz in a Redbook. Otherwise, severe aliasing artifacts will occur in ADC. If as close as possible to 'Brickwall' filters are used for this, a much bigger increase of the time smearing problems in the frecuencies below will happen. If instead a gentler filter slope is used, either you will have signals remaining beyond that NF and won't eliminate those aliasing problems you are trying to avoid, or, by displacing the filter to a lower frequency, you will start having a poor high frecuency response in the audible band , as that NF in Redbook is quite close to the audible limit. And yet, still having phase problems to some degree. Every ADC process must balance these opposing problems, but it is not possible to get rid of them completely. One of the reasons, according Meridian, why low sampling digital sounds harsh compared to analog sources.

Among the basic premises of MQA, one of the reasons why HD files (any high resolution file, not MQA only) sound better -if done properly-, is because that NF is displaced to a higher frequency. Then much gentler analog filters can be used to cut the incoming signal prior to quatization, without reaching the audible band, and so, preserving the phase coherence between fundamentals and harmonics in that audible band, while at the same time avoiding those aliasing problems of Redbook (at least, close to the audible region). There are of course other reasons that justify a higher sampling (like the better impulse response you get the higher your sample is), but I mentioned the issue because of this rather simple fact of avoiding the effects of *analog* filters in the audible band.

Then, if that explanation satisfies you, perhaps you may address the other "aberrations" of my previous post that you were commenting. It would be helpful for myself too.

It is true that filter implementations measure differently in the audible band.
The answer is to master in HD-Audio, high bitrate and bit-depth. Then optionally deliver at 44.1.

If HD-Audio sound better leave it alone.
Don't lower the sample rate, lower the dynamic range, apply a poor digital filter, and turn on a little blue light.
Many audiophiles chose different filters, SOX and such, who is MQA to limit this choice, especially, when their filter measures poorly.
Slow filters are not phase coherent so it is a bit rich to force the very problem, MQA proports to fix.
I am sorry, there is sufficient evidence that conclude, the MQA is predatory and lies to make money.
Pointing to other truths does not lend veracity to their huckstering.

If Audiophiles are access to both an MQA (down-sampled master) and the actual HD-Audio master, which one would they chose?
For an answer, I went to HDTracks, I found 3 tracks that seem to be samples.
This sometimes offers multiple resolutions and different price points, but for some reason MQA is not one.

- Rich

 

[KeithPhantom]

do you know an analog brickwall filter that doesn't mess the phases of the incoming signal?

As @mansr said, oversample and digital filter. This is also the same I’ve been telling you.

As this is problem that can't be solved perfectly, it think it is unavoidable that some degree of aliasing is present in Redbook files, because a tail of signal trespass the NF limit. It is also a obvious that, as ADC is trying to make this tail as short as possible, very steep filters are in fact applied,

It can be solved practically perfectly, by oversampling and the use of non-mathematically-ideal filters that we currently have. They cut enough ultrasonics so aliasing or resulting IMD is a non-issue since the level of “aliased” signals you are proposing is too low in the first place, even though there’s no aliasing with equipment using compliant filters to the sampling theorem (most of the market).

That's why HR files, having this problem shifted to a much higher frequency and with much more room to avoid steep filters, DO sound better (or at least have cleaner information stored if we are not able listen that difference [and we do, according Meridian]). And this is one of the foundational elements in MQA's theory.

You do not need MQA for this, technology nowadays already solved these issues. “Shifting frequencies” is already done by oversampling that pretty much all DS DACs and ADCs do. And about filters sounding better, you better include evidence for that claim.

 

[Tks]

There is no "double standard" as there is a significant difference between the tests of low-bit-rate codcs that I discussed and GoldenOne's test of MQA. I used "legal" test-tone signals - legal in the sense that they were appropriately band-limited and were at levels well below 0dBFS. The differences between how the MP3 and AAC codecs handled the test tones at lower and higher bit rates was therefore meaningful.

By contrast, GoldenOne's high-level ultrasonic test tones were not "legal," in that they didn't conform to the >20kHz spectral space typical of real music recordings. They therefore "broke" the encoder. See Amir's postings on this to which I was responding.

BTW: to make it easier for readers to access Stereophile's coverage of MQA since 2014, we have created an article category with all the relevant links: https://www.stereophile.com/category/mqa

John Atkinson
Technical Editor, Stereophile

I'm surprise you tolerate all you do considering if you had "ease of access" to the encoder, we wouldnt have to contend with much of the mess in the first place.

 

[Tks]

Not sure if I'm making it (I'm basing this comment precisely because of N-S in fact).

The only way for a Redbook to have a flat FR up to 20Khz is to apply filters above that 20 khz. The space left for that before reaching Nyquist Frequency (NF) is 20Khz-22Khz. In that space of less than 1/3 octave, the only possibility for the filter not to go beyond the NF (that is: to reach 0 db @ 22 Khz) is to be of brickwall- type, ie: of some 300 db/octave or so! (remind: most people here seems to believe that MQA need to handle the full 16bit amplitude even at 20 kHz: 96 db, as they are horrified that they can't reconstruct a square wave in the test tones). Besides the fact that this (a perfect brickwall filter in analog domain) is quite difficult if not impossible, do you know an analog brickwall filter that doesn't mess the phases of the incoming signal? The steeper the filter is, the more it shift phases of the incoming signal. Or we are going to question that, also?

As this is problem that can't be solved perfectly, it think it is unavoidable that some degree of aliasing is present in Redbook files, because a tail of signal trespass the NF limit. It is also a obvious that, as ADC is trying to make this tail as short as possible, very steep filters are in fact applied, and so, phases of the harmonics in the PCM has some (big, according Meridian) anomalities embedded in it.

That's why HR files, having this problem shifted to a much higher frequency and with much more room to avoid steep filters, DO sound better (or at least have cleaner information stored if we are not able listen that difference [and we do, according Meridian]). And this is one of the foundational elements in MQA's theory.

People use AA analogue filters for 44.1kHz? Who does this??

 

[mansr]

People use AA analogue filters for 44.1kHz? Who does this??

Straw men and red herrings.

 

[samsa]

The differences between how the MP3 and AAC codecs handled the test tones at lower and higher bit rates was therefore meaningful.

By contrast, GoldenOne's high-level ultrasonic test tones were not "legal," in that they didn't conform to the >20kHz spectral space typical of real music recordings. They therefore "broke" the encoder.

Not all of his tests fit that description. For instance, here was the result of encoding a perfectly legal 1kHz sine tone at -60dB in a 24/88.2 master:

How is that not a valid test of the MQA encoder?

And would you say that the MQA encoder "passed" the test?

 

[KeithPhantom]

Not all of his tests fit that description. For instance, here was the result of encoding a perfectly legal 1kHz sine tone at -60dB in a 24/88.2 master:

How is that not a valid test of the MQA encoder?

And would you say that the MQA encoder "passed" the test?

Is that some kind of noise-shaping? Because if that is a valid test, the encoder went bonkers there.

 

[samsa]

Is that some kind of noise-shaping?

One presumes that was the intention.

Remember that, after the steganography, 24/44.1 MQA has an effective bit-depth of 16 bits. To get a noise-floor below -100dB, you need some noise-shaping.

Because if that is a valid test, the encoder went bonkers there.

Indeed it did. Which is utterly astonishing.

 

[mieswall]

Okay, so I have been testing Tidal vs. Gobuz here couple of days.....comparison is very easy cause apps are in the same device........presentation is somehow thinner in Tidal, I dont know if it is the dynamics or what.......bass is more refined and stronger in gobuz, better sense of space also......sound wraps around your head in headphones, you dive deeper.......hihats are more real.........gobuz is closer to the sound of cd (which might funilly still win, havent tested).......tidals app is so far nicer , really like how it suggest artist and songs.....gobuz does this too, but so far not as well for my opinion..........by the way sound quality might get even higher if you download songs to the device, then u are not streaming over the net in real time, not sure about that yet.........I can hear difference with bluetooth headphones also, its not night and day ofcourse, but it is there...........So tidal interface might be tad better and faster ( probably because files are packed, which the can be heard) but it is not enough a reason for me to go back for now. Let’s keep testing.

I guess you comparison is with MQA rendered by hardware (way better than with the Tidal app alone). Do you have a MQA DAC, and are you sure it is rendering by hardware? (in a Mac, check Audio MIDI setup, select the highest format supported by your DAC; and in Tidal check that "Using Exclusive mode", and "Passthrough MQA" are both enabled for that DAC). Then try to compare files that turn on the blue light in your DAC, and preferably, of 96K or higher (in my case this sampling is not seen in Tidal or Midi Setup, that automatically changes to 44.1 when playing MQA's ; I can only check it at the screen of the DAC).

• I would suggest recent ECM files for comparison, most at 96K, but even those at 48K sound very good (I don't know if they are available in QoBuz). An older one that sounds good is Solstice, by Towner/Garbarek/Weber/Christensen: https://tidal.com/browse/album/77645332
• One ECM of low sampling, but that still sounds much better in MQA is Bobo Stenson's "Cantado": https://tidal.com/browse/album/79156357
• Another easy example that should be available in both is the Twin Peaks Soundtrack (MQA 192K): https://tidal.com/browse/artist/3792417. This record is available both in standard lossless and master in Tidal.
• Or this one, really amazing: 2L Audiophile Recordings, "The Nordic Sound". Most tracks @352K. https://tidal.com/browse/artist/4945930. If you can't find that one, any record from them sounds OK.Most of them classical music.

 

[Tks]

Straw men and red herrings.

Another part where he talks about the impossibility of having a true brick wall filter and that somehow being a problem in some appreciable or practical sense. Can you imagine telling this to Rob Watts after he put out his M-Scaler (while ridiculous) has as close to a brick wall filter as perfect as id ever care to have out of a dedicated device that wasnt using general software. I wonder what he would say to building an MQA-Scaler so OP wouldnt be able to make videos showing MQA encoder is polluting the audible band and things of that nature.

As an aside,

Meiswall also mentions 300dB/octave. What a quaint number that is for anyone remotely aware of Robs infamous ability. Rob Watts would definitely have some words to say about that seeing as how he could hear artifacts even down that low with his own ears. Maybe meiswall can also do that?

 

[Oukkidoukki]

I guess you comparison is with MQA rendered by hardware (way better than with the Tidal app alone). Do you have a MQA DAC, and are you sure it is rendering by hardware? (in a Mac, check Audio MIDI setup, select the highest format supported by your DAC; and in Tidal check that "Using Exclusive mode", and "Passthrough MQA" are both enabled for that DAC). Then try to compare files that turn on the blue light in your DAC, and preferably, of 96K or higher (in my case this sampling is not seen in Tidal or Midi Setup, that automatically changes to 44.1 when playing MQA's ; I can only check it at the screen of the DAC).

• I would suggest recent ECM files for comparison, most at 96K, but even those at 48K sound very good (I don't know if they are available in QoBuz). An older one that sounds good is Solstice, by Towner/Garbarek/Weber/Christensen
• Another easy example that should be available in both is the Twin Peaks Soundtrack (MQA 192K): https://tidal.com/browse/artist/3792417. This record is available in standard lossless and master in Tidal.
• Or this one, really amazing: 2L Audiophile Recordings, "The Nordic Sound". Most tracks @352K. https://tidal.com/browse/artist/4945930. If you can't find that one, any record from them sounds OK.Most of them classical music.

That is a good point, in that sense my test is not valid. I have smsl su 8, so no mqa. Exlusive mode is on on tidal on pc.....but.......I have streamed thru many sources.....android usb out, pc usb out, ipad 3.5 mm ear out, bluetooth......for example ipad ear out is 16bit 44 khz and difference is there too , even though both come out 16 bit ( no matter if it is hi res in app).....overall impression is that gobuz sounds better.........maybe somebody can compare with mqa dac.........edit. It is quite crazy idea that we need mqa dac to hifi sound.....
throw all these millions of devices to trash, stupid in another level...

 

[Jmsent]

Having read through all 34 pages of this thread, I was reminded of a Facebook post that Bruno Putzeys - of Hypex, PuriFi, and Kii-3 loudspeaker fame - wrote way back in 2017 regarding MQA. I don't know if it's been posted on ASR before, but it seems an appropriate time to bring it back up:

"This isn't a prelude to suddenly becoming active on FB but I felt I had to share this.
Yesterday there was an AES session on mastering for high resolution (whatever that is) whose highlight was a talk about the state of the loudness war, why we're still fighting it and what the final arrival of on-by-default loudness normalisation on streaming services means for mastering. It also contained a two-pronged campaign piece for MQA. During it, every classical misconception and canard about digital audio was trotted out in an amazingly short time. Interaural timing resolution, check. Pictures showing staircase waveforms, check. That old chestnut about the ear beating the Fourier uncertainty (the acoustical equivalent of saying that human observers are able to beat Heisenberg's uncertainty principle), right there.
At the end of the talk I got up to ask a scathing question and spectacularly fumbled my attack*. So for those who were wondering what I was on about, here goes. A filtering operation is a convolution of two waveforms. One is the impulse response of the filter (aka the "kernel"), the other is the signal.
A word that high res proponents of any stripe love is "blurring". The convolution point of view shows that as the "kernel" blurs the signal, so the signal blurs the kernel. As Stuart's spectral plots showed, an audio signal is a much smoother waveform than the kernel so in reality guess who's really blurring whom. And if there's no spectral energy left above the noise floor at the frequency where the filter has ring tails, the ring tails are below the noise floor too.
A second question, which I didn't even get to ask, was about the impulse response of MQA's decimation and upsampling chain as it is shown in the slide presentation. MQA's take on those filters famously allows for aliasing, so how does one even define "the" impulse response of that signal chain when its actual shape depends on when exactly it happens relative to the sampling clock (it's not time invariant). I mentioned this to my friend Bob Katz who countered "but what if there isn't any aliasing" (meaning what if no signal is present in the region that folds down). Well yes, that's the saving grace. The signal filters the kernel rather than vice versa and the shape of the transition band doesn't matter if it is in a region where there is no signal.
These folk are trying to have their cake and eat it. Either aliasing doesn't matter because there is no signal in the transition band and then the precise shape of the transition band doesn't matter either (ie the ring tails have no conceivable manifestation) or the absence of ring tails is critical because there is signal in that region and then the aliasing will result in audible components that fly in the face of MQA's transparency claims.
Doesn't that just sound like the arguments DSD folks used to make? The requirement for 100kHz bandwidth was made based on the assumption that content above 20k had an audible impact whereas the supersonic noise was excused on the grounds that it wasn't audible. What gives?
Meanwhile I'm happy to do speakers. You wouldn't believe how much impact speakers have on replay fidelity."
________
* Oh hang on, actually I started by asking if besides speculations about neuroscience and physics they had actual controlled listening trials to back their story up. Bob Stuart replied that all listening tests so far were working experiences with engineers in their studios but that no scientific listening tests have been done so far. That doesn't surprise any of us cynics but it is an astonishing admission from the man himself. Mhm, I can just see the headlines. "No Scientific Tests Were Done, Says MQA Founder".

 

[RichB]

I examine the issues that are faced by A/D converters in this article I wrote for Stereophile in 2018:
https://www.stereophile.com/content/zen-art-ad-conversion

John Atkinson
Technical Editor, Stereophile

I am no expert but, my takeaway from your article were found in these quotes:

With rare exceptions, the ringing at the sample rate's Nyquist frequency is ubiquitous with A/D converters. However, it's difficult to see why this should matter. While there will be some few listeners who hear a tone at 22.05kHz with CD data, no one will hear a 48kHz tone with data sampled at 96kHz, or a 96kHz tone with 192kHz data.

I believe "temporal blur" or "energy smear" is being used to discuss pre-ringing at Nyquist.

Keith Howard investigated the "energy smear" exhibited by digital filters in January 2006, writing that "the time-domain performance of anti-alias and reconstruction filters has increasingly been blamed for CD's residual failings." However, he concluded that the listening tests described in that article showed that the energy smear "seemed surprisingly reluctant to show its face"; only in one extreme case—a filter in which all the ringing occurred before the impulse—did this energy smear prove consistently audible in listening tests.

There does not appear to be real-word filters that places all the ringing before the impulse.
Then the article moves on from "temporal blur" and energy smear to "confusion". This seems to be following the MQA line.

Nevertheless, I understand that the ear/brain acts as a detector of wavefront arrivals, and that the pre-ringing of an acausal digital filter causes confusion: the initial onset of the ringing and the arrival of the maximum energy peak are incorrectly interpreted as two separate events rather than as one—as implied by Bob Stuart in his June 2018 interview with Jim Austin.

This might be one reason listeners prefer the sound of recordings made with higher sample rates (footnote 6). With 96kHz data, the time delay between the beginning of the sinc-function envelope and the maximum energy peak will be less than half what it is with CD data, and with 192kHz data it will be less than one-fourth that duration. Each time the sample rate is doubled, the time delay—and thus the confusion—will be half what it was before.

It is also possible the preference for HD-Audio could be also be FOMO (Fear Of Missing Out).

Is the concern about pre-ringing centered about 22.5 kHz pre-ringing, where the signal is extremely low and I suspect no one on this thread can hear?

I have seen measurements showing phase shift in the audible range can occur with minimum phase filters.
That seems to be a more appropriate discussion to start using terms like "energy smear", "temporal blur", and "confusion".
Where am I going wrong?

- Rich

 

[NTK]

I believe "temporal blur" or "energy smear" is being used to discuss pre-ringing at Nyquist.

Keith Howard investigated the "energy smear" exhibited by digital filters in January 2006, writing that "the time-domain performance of anti-alias and reconstruction filters has increasingly been blamed for CD's residual failings." However, he concluded that the listening tests described in that article showed that the energy smear "seemed surprisingly reluctant to show its face"; only in one extreme case—a filter in which all the ringing occurred before the impulse—did this energy smear prove consistently audible in listening tests.

This is from page 2 of Keith Howard's 2006 article in Stereophile in your link. It showed a complete misunderstanding on the sampling theory (i.e. how band-limiting works). Try band-limit an impulse to 22.05 kHz, and look at it in an oscilloscope. (Or look at the impulse response of an anti-alias filter.) When sampled, it will not give you only one single pulse. When you have only one single pulse in the digital samples, it means and it is the sinc waveform.

 

[KeithPhantom]

As @mansr said, oversample and digital filter. This is also the same I’ve been telling you.

About my own post. I think you still need an analog filter before the ADC stage. If I’m not right feel free to correct.

 

[laurelkurt]

Sorry for the attitude but I was just confirming your contentions.

It's Ok. I'm not crushed or anything. I'm here to learn what I can, which would be basic gear functions at best. I stay out of technical debates, but ask questions if something comes off as odd.

 

[mansr]

About my own post. I think you still need an analog filter before the ADC stage. If I’m not right feel free to correct.

Theoretically, yes, you need a filter to remove anything above half the sample rate. With a typical sigma-delta ADC, this is well over 1 MHz. A microphone feed used for music recording won't have any content at those frequencies in the first place, so there is nothing for the filter to do. Another way of seeing it is that the various bits of circuitry and cabling in front of the ADC already perform the duty without the need for an explicit filter.

 

[KeithPhantom]

Theoretically, yes, you need a filter to remove anything above half the sample rate. With a typical sigma-delta ADC, this is well over 1 MHz. A microphone feed used for music recording won't have any content at those frequencies in the first place, so there is nothing for the filter to do. Another way of seeing it is that the various bits of circuitry and cabling in front of the ADC already perform the duty without the need for an explicit filter.

So this means that if there are ultrasonics (realistically, no more than 24 kHz) in the recording, they would legal in the analog stage but would have to be filtered in the digital output?

Another way to ask would be if Sigma Delta ADCs sample at an oversampling rate and then use digital filters to output the desired sample rate? My knowledge of Sigma Delta in ADCs is a bit limited, thus my question.