• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). Come here to have fun, be ready to be teased and not take online life too seriously. We now measure and review equipment for free! Click here for details.

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

Status
Not open for further replies.

GoldenOne

Active Member
Joined
Jun 25, 2019
Messages
201
Likes
1,394
EDIT:
UPDATED with most current information.



Introduction
MQA is a proprietary music compression solution that uses the FLAC container. It requires a license fee for manufacturers to implement it, and due to the lack of transparency from MQA, there has been relatively little conclusive information about exactly HOW it works. (Or IF it works).
I was frustrated with this, and despite the fact that there is excellent testing that has been done by Archimago, MansR, John Atkinson and others, I wanted to get something done quite conclusively, to get some files encoded in MQA and published on Tidal, so that I could compare them to the original masters.

My tracks were published using MQA's recommended method for indie artists. And the tracks, including original masters, MQA encoded 44.1khz versions, and tidal 'hifi' versions are available here: https://mega.nz/file/S6AFQYQS#H5wK-6m6V7smG-TKdxxRVEfTOVE8vgxVv7Tiq56FVjQ
Feel free to have a look at them and poke around yourself.
I've included analog recordings of a "Full Decode", outputting from an ifi iDSD Diablo into an RME ADI-2 Pro FS R ADC at 24 bit 768khz.

If you have 40 minutes spare, I would recommend watching the video as I explain a little more about the context surrounding some of these tests, MQA's response, and other things.
But if you just want the hard data, read on.

I did contact MQA prior to posting the video (and this post), but their reaction was to immediately have my tracks removed from Tidal, and to have the publisher remove all my content.
I posted here, APS and Head-Fi to encourage people to search for and confirm the existence of the tracks, so that MQA could not later claim they never existed.
Only after I made these posts did I receive an email response which i'll discuss at the end of this post.





Test components
I wanted to get test files such as an impulse response, white noise and -60dB dynamic range test sine encoded in MQA.
When I attempted to simply submit these alone as "Alternative Electronic" tracks, I was told by the publisher that "The MQA Encoder was unable to encode the file".
I therefore assume that MQA has detection in place to ensure no one gets any test files encoded.
To my knowledge the Chord M-Scaler has something similar for impulse response detection.

So, I put together a basic acoustic track, and hid the impulse response, white noise, square wave, 32 tone test and full RightMark Audio Analyzer test sequence inside it with a few seconds of digital silence either side. As well as the musical content itself of course.
I did publish both dithered and non-dithered versions, and both exhibited the same issues that i'm about to show.



For the 88.2khz test I included a few changes.
- Test components such as impulse, whitenoise, RMAA test sequence were adjusted for 88.2khz, so white noise goes up to 44.1khz and impulse is one 88.2khz sample etc
- Added a 35khz+40khz sine, to test if this is removed or retained during MQA Unfolding
- Reduced duration of digital silence at end of file


I did also create another 88.2khz track with the amount of ultrasonic content drastically reduced, and no white noise. But MQA removed all of my content from Tidal before this went live and so I was not able to test this.

Additionally, i'd like to clarify:
This is NOT an attack on any manufacturer that has incorporated MQA into their products.. This post is intended to inform you about the issues and false claims surrounding MQA. There are a huge number of fantastic products which happen to have MQA support. This is about MQA specifically, not the products that support it.

Comparing the 44.1khz files:
The first thing I wanted to test was whether the 44.1khz master track was lossless, both on other platforms such as Deezer, and on Tidal.

For this I used the tool 'DeltaWave' on the Original 44.1khz master and Deezer version. Absolutely bitperfect. Same file.




However the Tidal version was NOT the same




Immediately it was clear that there was more high-frequency noise, and the MQA version actually generated clipping warnings (see top of image) whereas the original did not. This also seemed to be a common theme among commercial tracks I tested. MQA was consistently louder and would clip more often.

I don't actually think this is due to necessarily an overall volume increase though, but simply that even in the 44.1khz file, MQA adds what looks like minimum phase resampling artifacts. This causes some portions of the track to appear louder, transients to 'overshoot' and be more likely to clip.

This is the impulse test in the original file, one sample with digital silence surrounding it (note adobe audition adds the interpolation line hence the squiggle, but you can see the samples):




This was that same portion in the MQA 44.1khz file (time-scale is kept the same):



Not only is it showing minimum phase, and non-linear ringing characteristics, the ringing also does not decay linearly.
Square wave also looked....odd

Original:





MQA 44.1khz:




Lastly, the high frequency noise was not simply dither. It does look like MQA is adding its own dithering, as in the 24 bit un-dithered file I provided the noise floor had been raised, even in silent portions of the tracks.
However the noise rapidly increases in the higher frequency area.

This is a -60dB 1khz sine, used for testing dynamic range. But the same noise pattern exists even in areas of the track that SHOULD be silent.

Original:





MQA 44.1khz:






So no, the 44.1khz file is NOT lossless at all. And shows evidence of compression/resampling, even though the file sample rate has not changed. (Similar effects are seen when compressing to MP3).
But, MQA offers no file size advantage here. In fact the 44.1khz MQA file is a few megabytes bigger than the FLAC master I provided (WAV is bigger obviously).


The 88.2khz file

Next, let's look at the 88.2khz file.
The version on deezer had simply been resampled to 44.1khz. Any >22.05khz content was just removed, and the impulse response showed the resampling but was phase linear. So not lossless, but then Deezer doesn't offer native HiRes, and so a resampled but otherwise fine version is what we'd expect.

The MQA version was.....well a picture says a thousand words:



There is a HUGE amount of high-frequency noise. So much so that for our dynamic range test this is now actually 20dB ABOVE the signal level, with a further increase in noise around the fundamental as well.





The impulse response was near impossible to make out from the noise so isn't shown in this post, but files are linked in the intro if you want to have a look yourselves.

Additionally, any >22.05khz content had been aliased back down into the audible band at fairly audible levels. With my 35khz and 40khz sines now at 4.1khz and 9.1khz. And you can quite clearly hear them being played. They are absolutely NOT 'below the noise floor' as MQA claims.
There was also notable added artifacting in the audible band.






So no, the claim that a hires file "Folded" to 44.1khz can just be played back on a normal DAC with no issue at all is certainly not true. Content is aliased into the audible band and not at acceptable levels.

I did prepare another track to test two things
Firstly, this HF band of noise is not present in many MQA files that are sourced from HiRes masters. Sam Smith's "Too good at goodbyes" is an example of a track where there is a native 88.2khz and MQA version sourced from an 88.2khz master available. It does not have this excessive HF noise. BUT, looking at the original 88.2khz master, you can see that the >22.05khz content was cut anyway. Therefore I suspected that the level of noise added by MQA is proportional to the amount of ultrasonic content it has to fold/reconstruct.

Secondly, I wanted to test what level MQA would encode lower level ultrasonic content down to. Say a signal at -60dB instead of -10dB.

I had submitted this to the publisher, but unfortunately MQA had removed all of my content from Tidal before this track went live and I was never able to test it.
If MQA sends me this file, I will update this post with findings. But I doubt that is going to happen.


Unfolding 44.1khz
So, what happens when we unfold either of these files?

Firstly, I found it curious that even the file where i'd given a 44.1khz master authenticated as "MQA Studio" on the DAC, not just the normal MQA green/magenta light.
Why is MQA unfolding this at all? There was never any ultrasonic content in the master to 'restore'.
Other issues with MQA's "Authentication" have been demonstrated by @FredericV here:


I think it's fair to say MQA's "Authentication" authenticates nothing.
The file is NOT the same as my master, you can alter and chop off large portions of it and still have it authenticate, and it was also never from a high-resolution master.


This imaging is a result of MQA's "Leaky" filter. An issue which has been discussed both by @Archimago in his excellent write-up here:

And by @The Computer Audiophile in his 2018 RMAF talk:

In that talk by the way, MQA handwaves away concerns and implies that the information is simply biased, and cannot be trusted unless Chris reveals his source's real name.
He quite rightly said that ANYONE can demonstrate this issue. And if you'd like to do just that, i've written a post how to do exactly that here:

This post guides you how to test the 'first unfold'.
Mans Rullard has also created a tool to add MQA flagging to any MQA file to trick a DAC into using it's internal MQA renderer filter if you'd like to test the 2nd/3rd unfold "renderer". I've added a .zip for a working windows version to this post, called "MQATools.zip". Simply extract, open up powershell and type:

./mqbgen.exe inputfilename.flac outputfilename.wav
Easiest way is put the input file in the same folder first.
@Miska has also posted a graph of the renderer filter in that thread.

Using the whitenoise in my file, this was the response I got from the 44.1khz file by doing an internal Full Decode on the iDSD Diablo (with HQPlayer Sinc-L in white to show where noise is due to device limitation)




Fun fact: The ifi GTO filter is basically the same as the MQA Full Decode:



Anyway, getting back on-topic, the 44.1khz file simply exhibited imaging problems due to the 'leaky' filter. There was no high frequency content 'restored' because it was never there in the first place. Neither the stock or unfolded 44.1khz files were "lossless".

More concerning, Tidal actually did not offer a lossless version of my track. And i've since found that whilst a track/album NOT marked as "master", it will genuinely stream lossless. For any track/album where it IS marked as "master", the 'hifi' setting simply streams the MQA file but without MQA flagging and does not unfold it. It is the same file!!




This means that for any track marked as "Master", Tidal no longer offers a lossless version, you ONLY have compressed, or lossy MQA.
Tidal also shows absolutely 0 indication about what the original master sample rate was, even though this info is included in the file and services such as roon will display it.
This means that customers have absolutely no idea what they're getting, whether it is from a true Hi-Res master or just poor upsampling, and more to the point, the stock 44.1khz/HiFi version is NOT LOSSLESS. You literally CANNOT avoid MQA for a track marked as master unless you opt for compressed.
Some releases have a separate listing which is not marked as MQA, but this is not universal unfortunately.

The impulse response also now had even wonkier stuff going on which I really don't know how best to describe:



Lastly, a "Full Decode" of the file just results in upward imaging further:



Unfolding 88.2khz
So what about unfolding the 88.2khz version? Does it actually restore high frequency content? Are their claims true?
Well...actually yes....sort of....










With the first unfold done, as you can see above, it DOES put back the high frequency content. However that massive band of noise is still present. Centred on about 25khz.
The audible band has had the overall level of noise reduced compared to the 44.1khz file, but there is still noise present. Additionally, in the places where MQA is needing to 'restore' high frequency content, it leaves audible band artifacts which are not present in either the original master, or actually the 44.1khz MQA file either. And do not exist in parts of the track where there was no ultrasonic content.
This means we effectively have a reduction in dynamic range/added noise whenever MQA is 'restoring' content. Both in the audible and ultrasonic bands.

So to my surprise, yes, MQA does restore ultrasonic content, however it is still quite frankly plagued with issues and the result is still nothing close to my original master.
Once a 'Full Decode' was done, these problems still exist and we just see upward imaging once again.




So:

- MQA is not lossless in either stock or unfolded form. Regardless of original file sample rate.
- Most MQA releases are simply being upsampled, and are NOT sourced from high resolution masters.
- MQA aliases high frequency content down into the audible range with minimal attenuation.
- MQA does put back high frequency content, and remove the previously aliased content when unfolded, but leaves distortion and noise behind.
- Tidal no longer offers true lossless streaming for any track marked as "Master" that I could find. You either have compressed, or lossy MQA. The lossless FLAC is not available and you must use an alternative service such as Qobuz or Deezer.
- MQA 'authentication' and the blue light gives no guarantee of source sample rate or lack of alteration. And does not guarantee that the sound is the same as was in the studio
- A huge number, presumably most, of MQA releases are just upsampled and altered 44.1khz. Not Hi-Res
- MQA like for like offers no file size advantage, and is actually bigger than 44.1khz FLAC. Though is smaller than Hi-Res flac.



Opinion

Now, with the testing out of the way, i'm going to give my own opinions about this situation, MQA in general, and why I did this testing.
The following is my own view, not necessarily fact:



I am an audiophile. I love the gear, I love the music, and I love when a technological advancement brings an improvement to our listening experience.
But anyone in this hobby will tell you that there is a LOT of snakeoil about.

There are plenty of products that promise a better experience and claim to be the best way of doing whatever it is they're doing. But there are two key categories.



- Those that have evidence or can be tested by third parties
- Those that are cannot



There are people like Rob Watts of Chord, who is quite vocal about the fact that he believes his WTA filter design is the optimal method for playing back 44.1khz audio. And not only is evidence given for this, you can test it yourself.
Anyone with a chord dac or a chord M-Scaler is free to play and record any music or test signals they wish, and Rob Watts has also been very responsive on places such as head-fi, writing up detailed posts about how and why his design works.

He isn't going to hand over the filter coefficients, but he'll answer enough questions and provide evidence for you to make an informed decision as to whether you agree and would like to purchase his product.
Same goes for @Miska and his excellent HQPlayer software, which literally has a version that provides direct file output, which you can test with upto 60 sec audio files for free.
Miska is super active on APS and on roon forums answering people's questions about both his own product and the area of digital reconstruction and playback in general.


Similar story with DSD. It is open, anyone can encode and play back files using it, and it is completely able to be poked, prodded, explored, and have the pros and cons discussed.



But then there are companies like MQA.
Not only has MQA has not provided evidence. They prevent you from gathering your own.
You cannot encode your own files in MQA or record the final output of a full decoder, it just operates on a "trust us its better" approach.
Their marketing claims are vague and change and they get disproven.
They have released information and done some things that are genuinely fascinating. MQA's "white glove" treatment, correcting for problems in recording equipment, is truly fantastic and giving the "fairytales" article a read is highly recommended. But the way it is presented makes people believe that this is how MQA's core product works. When it isn't.

The implication is that all MQA releases are better than FLAC, when the majority of it is just upsampled 44.1.
Bob Stuart's blog provides an explanation for how MQA folds high frequency info, which could never work with a 16 bit file as it wouldn't have enough dynamic range.

When issues are presented to them they dodge the question and try to divert the conversation.

And all the while, more and more music content is being replaced with MQA, more and more hardware is supporting MQA, and it continues to spread, with the supposed benefits being unproven and its problems only steadily being exposed as people find ways around the testing roadblocks.

The reason hardware manufacturers are supporting MQA isn't because they genuinely think it works and is better than native hires. Its because you, the consumer, are demanding it.

Various companies such as Schiit, Linn, and even PS audio who's DACs do support MQA, have spoken out against MQA.

Paul from PS audio talks in a video about this, saying that it was added because so many people were demanding it, but he still doesn't like it.

It is a business decision, and regardless of MQA's technical performance, no one can dispute that it is an absolutely fantastic example of the power of marketing.

I like progress, I like advancements in allowing us to enjoy music all the more. But MQA has various demonstrable problems, none of which exist if you simply stream native hires flac or even just 44.1khz.
I cannot recommend ANYONE use MQA in any situation, and you should vote with your wallet, don't buy products that have costly MQA licensing fees tacked on, don't subscribe to platforms like Tidal that are removing lossless content. And demand that companies be honest and transparent about the products they sell.

If a company can't provide it's own evidence, and goes to great lengths to prevent anyone else testing it, that should probably raise a few red flags.



MQA's Response
Immediately after MQA removed my content from Tidal, I posted here and in a few other places to encourage people to go and find the tracks on Tidal or Roon. They could not be streamed, but could be seen. This was so MQA wouldn't later be able to claim they never existed, and there would be "witnesses" so to speak.

AFTER that happened, I received a response from MQA. Their full response can be read here: https://pastebin.com/2YfT6vfZ

I won't address every point as much of it is marketing speak. But there are a few i'd like to address:

- The first part discusses how MQA is more advanced than conventional approaches and intends to account for the conversion process at each end of the chain.
To which i'd respond with two things:

1) In that case why is Warner, who offers MQA for seemingly most of their releases, not using MQA for archival storage? Or anyone else for that matter?
2) Why was I never even asked during the publishing/MQA encoding process about what recording methods or hardware i'd used? I also used a fully MQA certified full decoder for output so that cannot be the issue. And digital recordings were made from Tidal or Roon's MQA core decoder.

More to the point, even if you DID have info about the source and DAC hardware components, how would you account for tracks where multiple ADCs, synths, recording methods and in-production editing methods were used? Hardly any music nowadays is produced in analog and then digitised, or digitally recorded using only one ADC model.

- "We don't understand your frustration about evaluating MQA."

My frustration is that no end to end analysis tools are available like they are for nearly EVERY other format. Yes we can listen to any MQA song, but we have no information about how its been altered, is it the same master, no way to check for placebo, or any way to objectively test with test files other than jumping through hoops like I had to do for this video/post.
Additionally, when someone like Chris Connaker, Mans Rullard or Archimago provides evidence, the question is dodged, and when I reach out to you, you attempt to censor me by removing all of my content.
THAT is my frustration. It isn't transparent in the slightest and MQA responds in a hostile manner whenever criticised.
If you truly stand by your claims, myself, and just about anyone else would be more than happy to work with you to conduct testing that both sides agree on to demonstrate your claims. But when all the evidence suggests your claims are false, and you do not allow proper testing or provide your own, there is absolutely no reason to believe you.

- "This [the MQA encoder] is not configured to deal with content where, for example, the statistics change mid‐song, or where the audio does not resemble natural sound."
Does that then imply that electronic music won't work with MQA? Or that tracks with a combination of 'natural' recording and synthesised parts won't work?
Right now none of the tracks in Tidal's top 10 are true analog recordings, and 8/10 are in MQA.
Maybe the 'studio tools' that are referenced in your response do help to overcome these issues, but again, there is no transparency. We have no idea what these tools are, who has access to them, which releases were made using them, and have no way to evaluate them.
Would you ever consider making a limited version of these available? Perhaps releasing a version that only allows for a max 10 second audio length. This would allow people to test properly.
A little transparency would go a LONG way. And saying that MQA is lossless, but only within certain vague and unspecified criteria, and lossless only using non-conventional evaluation methods, isn't exactly reassuring.

- In regards to my mention of the blue MQA authentication indicator they said: "The onus is on the submitter to check the content when it arrives in Tidal and confirm the sound."
This was NEVER mentioned during the publishing process. And doesn't address the fact that you can take an MQA file, throw away a third of it, and still have that light show up.
Plus, even the 44.1khz file I submitted both sounds different and is objectively different to the master I published.
And again this is the same for any other track where a high sample rate version and MQA version are available, they are not the same.

- The next part of their response goes on to explain how MQA adaptively identifies and responds to content in a track.
This again, implies that it cannot be lossless. If it is behaving differently to different types of recording and music then the result will not be consistent.
It might sound more 'natural' to some people, but sounding natural and being lossless to the original master are not the same thing.
If you wish to market an 'AI upsampling' or responsive/adaptive process then that's totally fine. But sounding subjectively better to some people does NOT make it lossless and is no excuse to make false marketing claims.

- They also discuss the files I had published.
I should firstly mention that the information and status messages they're providing here, I only ever received one of these, the one saying that the encoder was unable to encode the file. And that was only for the files where it was literally just a test file, an impulse response or square wave without any other content, which would likely not pass MQA's checks for this sort of content.
All the other error messages, I never received, I do not know if these are genuine, and given as MQA had the publisher remove all of my content I cannot go back and check.
Given as I cannot verify any of these, and even if they are true it doesn't address the fact that the encoder was unable to handle my files, I won't discuss this point further.

- In response to my concerns about added noise, MQA claims that this was my fault, as I didn't dither the files. Claiming this was a "Naive mistake".
No MQA, it wasn't. In fact this was something I had explicitly tested. The first track I submitted, called "Try again", DID have dither, and showed all of the issues mentioned in this video, in fact some to a worse extent. To give MQA the benefit of the doubt and check that the dithering wasn't adversely affecting anything, I published the next two tracks without dither.
So this is not my fault at all, and the only Naive mistake made was not looking at all three files I had published.

- In response to others and my description of the MQA upsampling filters as leaky, they claim this term is derogatory and inappropriate. And that the only alternative would be brickwall bandlimiting.
Firstly, I don't think that the filter itself has feelings, but if I have hurt them then I apologise. Secondly, perhaps the biggest advantage of genuine native hires audio is just this. It allows much more flexibility with filter design. Going from 44.1khz to 96khz source sample rate gives you over 13x more distance between the audible band and nyquist frequency.
So if the argument is that brickwall filters are bad, which to be clear, in many aspects I agree with, the solution isn't MQA, its to use native hires!

- In regards to aliasing they claim that this is simply because the levels of the signal were too high and it wouldn't occur in real music.
For this point, I'm quite happy to accept this might be the case. (Though I would say it still is concerning that the encoder wasn't able to handle even simple sines, and still invalidates the "Lossless" claim.)
In fact as I mentioned earlier in the post, i'd prepared another track with ultrasonic content at much lower levels to test what level this was aliased down at. Unfortunately you had my tracks removed and this never made it to Tidal.
If this claim is true, then send me the MQA encoded version of that file and i'd be happy to post here with updates.
Again, lack of transparency makes it difficult to address concerns.....

- They also say that every MQA file will tell you the sample-rate of the original audio.
And this is true, but my complaint wasn't that it didn't tell you at all, it was that it isn't clear in the slightest. Tidal shows no indication at all of original sample rate.
You have to use a program like roon to show this information. This means it is misleading consumers into believing that any MQA release is better than RedBook FLAC.

I conducted a survey, which got over 250 responses, and of those who said they thought MQA was a good thing, TWO THIRDS of them believed that not only was MQA better than redbook, it was the same or BETTER than 24 bit 192khz FLAC.....
This is a concerning statistic as it simply isn't true!

I understand this lack of transparency may be Tidal's fault not MQA directly, but it is the combination of Tidal's lack of transparency and MQA's marketing that is having the harmful effect of misleading consumers.

Lastly i'd also just like to say. Obviously this is not 'ideal conditions' testing. But given the evidence shown and MQA's response, quite frankly until they provide something concrete, there is just no reason to trust them at all.
I, Amir, and many others would be HAPPY to work with MQA to test various claims, but the testing done here is the best we can do with the tools available.
Until MQA becomes more transparent, the objective evidence from various sources suggests it is riddled with problems.




Further Reading:

Bob Stuart wanted to censor AS: https://audiophilestyle.com/forums/...porware/page/584/?tab=comments#comment-958011

How to test MQA's leaky filter: https://audiophilestyle.com/forums/topic/62669-how-to-demonstrate-mqas-leaky-filter/

2018 RMAF MQA talk by Chris Connaker:


Archimago's MQA article: https://audiophilestyle.com/ca/reviews/mqa-a-review-of-controversies-concerns-and-cautions-r701/

PS Audio vid on MQA:

Schiit on MQA: https://www.schiit.com/news/news/why-we-wont-be-supporting-mqa#:~:text=Today, Schiit Audio announced that,so that they may choose

Linn on MQA: https://web.archive.org/web/20201111211105/https://www.linn.co.uk/blog/mqa-is-bad-for-music

Neil Young removes albums from TIDAL: https://neilyoungarchives.com/news/1/article?id=Tidal-Misleading-Listeners

John Atkinson of stereophile confirms low-level distortion: https://audiophilestyle.com/forums/...porware/page/680/?tab=comments#comment-980420

GTO filter analysis: https://audiophilestyle.com/forums/topic/55951-ifi-foe-fum/#comments

Benchmark Audio's statement about MQA transient shifting: https://www.theabsolutesound.com/ar...lifier-and-dac3-b-digital-to-analog-converter

RealHD stops supporting MQA: https://www.realhd-audio.com/?p=7218

Archimago on transient smearing https://archimago.blogspot.com/2018/02/musingsmeasurements-on-blurring-and-why.html

Rob Watts on digital filters: https://www.head-fi.org/threads/watts-up.800264/page-134#post-15968970

Benchmark on MQA: https://benchmarkmedia.com/blogs/application_notes/163302855-is-mqa-doa

Meridian has doctored tests of hi-res in the past: https://audioinvestigations.blogspot.com/2015/07/the-great-debate-amir-vs-arny.html

Filter ringing (with a link to Mans/troll audio) https://audiophilestyle.com/forums/...porware/page/699/?tab=comments#comment-983918

Aliasing vs. imaging: https://audiophilestyle.com/forums/...porware/page/683/?tab=comments#comment-980885

Music aliasing distortion example: https://audiophilestyle.com/forums/...porware/page/683/?tab=comments#comment-980857

Linear phase vs. Minimum phase: https://troll-audio.com/articles/linear-and-minimum-phase/

Sox settings to duplicate an MQA filter: https://audiophilestyle.com/forums/...echnology/page/7/?tab=comments#comment-754014

JA's post showing signal aliasing: https://audiophilestyle.com/forums/topic/30381-mqa-is-vaporware/?do=findComment&comment=980401

mansr's original Dragonfly DAC analysis: https://audiophilestyle.com/forums/...analysis/page/20/?tab=comments#comment-690307

Why MQA's digital filters blur the music: https://audiophilestyle.com/forums/...porware/page/470/?tab=comments#comment-936690

A Survey of Musical Instrument Spectra to 102.4 KHz https://www.cco.caltech.edu/~boyk/spectra/spectra.htm

RealHD Audio on MQA: https://www.realhd-audio.com/?p=6517


EDIT:
ORIGINAL POST FOLLOWS.


I figured that given how aggressively Tidal has been expanding their use/incorporation of MQA, and there seems to be an awful lot of debate about whether or not MQA is good or lives up to the claims, and not much testing going on, (including lack of evidence from MQA themselves), I should try to remedy that.

TLDR: MQA isn't lossless, is arguably worse than normal flac, and is seemingly nothing more than a (quite effective) scheme to generate licensing fees. With the frustrating addition that if you are a Tidal user, even if you have no MQA dac, and use the "Hifi" streaming quality setting, MQA encoded/lossy files will still be served to you. And the only way to avoid that being to switch to Qobuz


This post is intended to answer test and answer a few questions about MQA, namely:

1) Are MQA releases the same master as non-MQA?

2) If you don't have an MQA dac, is standard FLAC and MQA-FLAC the same / does MQA provide a benefit even on a normal dac?

3) Is unfolded MQA lossless or as good as native HiRes?



This is normally quite tricky to test because MQA ensures that there are no native HiRes releases for tracks that are released in MQA on tidal. So you cannot directly compare them. However, there are a couple which seem to have slipped by.

Absofacto's "Thousand Peaces" for example has ONE of the songs in 96khz on qobuz (the rest are 44.1) and 88.2khz via MQA on tidal. I initially tested this, however it turned out that the Qobuz redbook and tidal redbook versions were different, meaning they are using different masters and could not be directly compared.

Answer 1: MQA/Masters SOMETIMES uses a different master source. Meaning the file formats themselves cannot be compared as the information itself is different. This is likely done to give the impression of sounding better even though it's nothing to do with the file format.



So then, we need a different test track/album. Sam Smith's "The thrill of it all" however was ideal. It has a native 24 bit 88.2khz version on qobuz as well as the standard 44.1khz release. And on tidal there is also a 44.1khz release and can be 'unfolded' to 88.2khz via MQA. Meaning we can compare identical sample rates.

The first thing to do was to check whether the Tidal and Qobuz redbook/non-MQA files were actually the same. ie: Are tidal and qobuz using the same master for the song. To do this I downloaded the Redbook 16 bit 44.1khz version from Qobuz, and then the same from the release on tidal that was not marked "Master".

Deltawave showed that these two files were 100% absolutely bit for bit identical. So we can conclude that Tidal and Qobuz are using the same master for the song. Perfect.

1616639359050.png


Next, I downloaded the "Master"/MQA release, but without any MQA unfolding. ie: keeping it as a non-MQA dac owner would be playing it. Both these files are 44.1khz, but are not the same. In fact they are only 0.43% bitperfect with a 40dB null (24 bit accuracy is 146dB) We can see that the master is clearly the same as the majority of the track is identical, but the MQA version has a significant amount of high frequency noise compared to the lossless FLAC.

meXPEfq.png


(Y axis is frequency, X axis is time. Green means that part is the same, purple/red means it is higher or lower in level and different from the original)

Answer 2: If you do not have an MQA DAC, MQA should be avoided, the content is NOT the same as the lossless original, and has more high frequency noise.


So then, now we need to see what happens if you unfold the MQA version to 88.2khz and compare it with the native 88.2khz version. I did this by using Roon, which has MQA decoding support, and recording bitperfect output, then comparing against the native hires 88.2khz version from qobuz.

Now things are really quite messy. The unfolded version differs significantly from the native hires, with again a lot more high frequency noise, as well as a band from about 11.5khz to 13.5khz where content differs a concerning amount in this specific instance.

UmjBeRE.png


Therefore

Answer 3: No, MQA is NOT lossless (a claim which MQA has recently removed from their marketing material), and even when unfolded does not match native HiRes content. I would love to test a full decoder/renderer, but MQA does not allow any "Full Unfolding" device to have a digital output. (Gee I wonder why that is, it'd sure be a shame if someone were to so easily be able to record and disprove the marketing claims.)

Additional arguments:

  • MQA is actually probably worse than native playback. MQA makes it basically impossible to obtain a "normal" and MQA version of the same hires file. BUT, Stereophile did manage to convince them to send an MQA encoded single-impulse file. Their testing showed three things:
    1 - Playing back an MQA encoded file on a non-MQA dac caused issues, and created an asymmetric impulse response.
    2 - Playing it back on an MQA capable dac, it was minimum phase, not linear.
    3 - Playing back a NORMAL, non-MQA encoded impulse response file, with the MQA filter turned on on the DAC, produced an IDENTICAL result to the MQA file, suggesting that MQA is nothing more than a basic minimum-phase upsampling filter in this situation, and absolutely nothing to do with the source file. https://www.stereophile.com/content/mqa-tested-part-1
    There is significant evidence from multiple third party sources to show that MQA has all sorts of problems. http://archimago.blogspot.com/2018/02/musingsmeasurements-on-blurring-and-why.html


  • MQA incurs an additional cost to you. You are paying for the licensing fees that are tacked on to products to get MQA support, and at every other step in the process. A good post from the manufacturer Linn is available here: https://www.linn.co.uk/blog/mqa-is-bad-for-music Given as we have now demonstrated that MQA is NOT a substitute for native HiRes content, its hard to argue that MQA is doing much more than charging you for a sub-par version of something you already had (native hires music). If you want the best quality, demand native hires releases, not licensed, closed-source, proprietary compression. Schiit audio has also spoken on it: https://www.schiit.com/news/news/why-we-wont-be-supporting-mqa


  • MQA IS NOT sourced from a HiRes master. Even if you are happy with it not being lossless, it is not actually even compressed from a HiRes source. Neil Young removed his music from tidal when after providing 44.1khz masters, Tidal suddenly released MQA versions, which would have been created simply by altering/upsampling the original. He did NOT provide them with HiRes masters to release in MQA, and you can read about this here: https://neilyoungarchives.com/news/1/article?id=Tidal-Misleading-Listeners
    "Tidal's master is a degredation of the original to make it fit in a box that collects royalties. That money ultimately is paid by listeners, I am not behind it. I am out of there. Gone. My masters are the original."
    MQA is at least in some situations simply an upsampled version with a licensing fee slapped on.....


  • There is ZERO proof of any of MQA's claims. There is absolutely zero evidence to support any of their marketing, claims that they can fit 24 bit 192khz content into a 16 bit 44.1khz file, and in fact, all objective evidence and testing so far (including this post) conclude that MQA's claims don't make sense at all. The claims they make would be VERY easy to demonstrate and prove if they were true....
    Most MQA content cannot be obtained in native HiRes anywhere. And they do not allow any "full unfolding" device to have a digital output to prevent anyone from recording or testing the result.


Thanks for reading, hopefully this helped some people!
 
Last edited by a moderator:

Helicopter

Major Contributor
Forum Donor
Joined
Aug 13, 2020
Messages
2,392
Likes
3,266
Location
Michigan
Nice work.

I already hated the idea of buying a more expensive DAC to use a proprietary format with licensing fees, when open source does it just as well, but now I understand the deep and vieled technical flaws in the format.

The only way it might be attractive to anyone is the lossy compression makes music cheaper to transmit.
 

Ron Texas

Major Contributor
Joined
Jun 10, 2018
Messages
4,036
Likes
5,214
Location
A mysterious place with no name.
In MQA some of the bits are used to do a lossy compression of high frequency information. On a CD that's 3 out of 13 bits. In a 24 bit format it's 8 to 11 bits. So, if you don't have an MQA DAC your MQA CD is 13 bits instead of 16 bits. The encoded 3 bits reproduce as low level noise. Don't buy MQA if you can avoid it.
 

scott wurcer

Major Contributor
Audio Luminary
Technical Expert
Joined
Apr 24, 2019
Messages
1,503
Likes
2,763
Do we need another MQA thread? The effort they put into preventing any verification of any of their claims should be enough reason to just ignore them. When supposed engineers lie to customers they get fired.
 

UKPI

Member
Joined
Mar 22, 2021
Messages
34
Likes
103
Location
South Korea
Do we need another MQA thread? The effort they put into preventing any verification of any of their claims should be enough reason to just ignore them. When supposed engineers lie to customers they get fired.
Unfortunately, ignoring them doesn't prevent disinformation from spreading. MQA is routinely mentioned as a hi-res audio format in other audiophile communities that I know, and ignoring it won't change that trend. The OP can be used as another evidence provided by an independent third party that MQA doesn't do anything beneficial, so I think this thread is more beneficial than 'just another MQA is bullsh*t' thread.
 

Lsc

Member
Joined
Feb 27, 2020
Messages
97
Likes
65
So if you have a MQA DAC like the Topping D90, then it’s all good?

I currently use Tidal desktop app to my processor and use the DAC from the processor.
 

voodooless

Major Contributor
Forum Donor
Joined
Jun 16, 2020
Messages
1,792
Likes
2,581
Location
Netherlands
So if you have a MQA DAC like the Topping D90, then it’s all good?

How do you reach that conclusion? The point is that it will never be “good” (as in perfect). The original signal is never preserved. Even worse a band of artifacts is in the audible range (if you are not to old ;) ). It seems to have varying effectiveness in restoring the HF spectrum. I wonder why? I’d love to see the waveform overlayed on the diff images to correlate.
 

abdo123

Major Contributor
Forum Donor
Joined
Nov 15, 2020
Messages
3,272
Likes
2,874
Location
Brussels, Belgium
Where are the double blind listening tests? How do we know that the differences you showed are actually audible?

Where are the MQA vs 320 kbps vorbis vs 256 kbps AAC ABX listening tests?

The conclusion of your comparisons are basically ‘everything we already knew about MQA is true’.
 

Lsc

Member
Joined
Feb 27, 2020
Messages
97
Likes
65
How do you reach that conclusion? The point is that it will never be “good” (as in perfect). The original signal is never preserved. Even worse a band of artifacts is in the audible range (if you are not to old ;) ). It seems to have varying effectiveness in restoring the HF spectrum. I wonder why? I’d love to see the waveform overlayed on the diff images to correlate.
Ok I just subscribed to qobuz. I’ll compare them.
 

TurbulentCalm

Member
Joined
Mar 18, 2021
Messages
82
Likes
190
Location
Australia
What is missed in this analysis is that the Tidal App does the first unfold in software. Therefore no DAC would ever be required to process the raw MQA/FLAC encoded data. I think you need to reconsider your findings.

you need to compare the actual unfolded bit stream if you are to provide an effective evaluation of the non-MQA DAC. How you do this though escapes me.

2L Audiophile Recordings: http://www.2l.no/hires/ has a range of hi-res tracks available for free download. Each track is provided in MQA, FLAC, DSD and DxD at different sample rates. You definitely could use these to compare audio formats.
 
OP
GoldenOne

GoldenOne

Active Member
Joined
Jun 25, 2019
Messages
201
Likes
1,394
What is missed in this analysis is that the Tidal App does the first unfold in software. Therefore no DAC would ever be required to process the raw MQA/FLAC encoded data. I think you need to reconsider your findings.

you need to compare the actual unfolded bit stream if you are to provide an effective evaluation of the non-MQA DAC. How you do this though escapes me.

2L Audiophile Recordings: http://www.2l.no/hires/ has a range of hi-res tracks available for free download. Each track is provided in MQA, FLAC, DSD and DxD at different sample rates. You definitely could use these to compare audio formats.
That is what the third comparison is. It is a comparison of the first unfold done by roon/tidal vs the native 88.2khz version.

As shown, its quite problematic
 

nimar

Active Member
Forum Donor
Joined
Jan 25, 2021
Messages
213
Likes
199
Location
Ontario, Canada
I don't believe MQA is manna from heaven, or the answer to audiophile dreams. More likely it's a way for the record industry to have tighter controls over the release of hi res music, while still holding on to the originals. Fine.

But the attempts at "proof" showing it to be inferior are poorly constructed and would be torn apart in any other area, if it wasn't for the fact that everyone loves to hate MQA.

In reverse order

There is ZERO proof of any of MQA's claims. There is absolutely zero evidence to support any of their marketing, claims that they can fit 24 bit 192khz content into a 16 bit 44.1khz file, and in fact, all objective evidence and testing so far (including this post) conclude that MQA's claims don't make sense at all. The claims they make would be VERY easy to demonstrate and prove if they were true....
Most MQA content cannot be obtained in native HiRes anywhere. And they do not allow any "full unfolding" device to have a digital output to prevent anyone from recording or testing the result.

Did any new format have to prove how it worked to the general public, maybe I missed that. People still pay a premium for DSD files, which are proven (its a fact) to be upsampled from the DXD (PCM) masters and we don't read about how its a scam every few weeks. I couldn't say what the motivation for restricting digital out is, but one could easily record the analogue out with an ADC and compare to a known like for like source such as 2L. Plenty of folk with such devices around, yet I've not seen this done?

MQA IS NOT sourced from a HiRes master. Even if you are happy with it not being lossless, it is not actually even compressed from a HiRes source. Neil Young removed his music from tidal when after providing 44.1khz masters, Tidal suddenly released MQA versions, which would have been created simply by altering/upsampling the original. He did NOT provide them with HiRes masters to release in MQA, and you can read about this here: https://neilyoungarchives.com/news/1/article?id=Tidal-Misleading-Listeners
"Tidal's master is a degredation of the original to make it fit in a box that collects royalties. That money ultimately is paid by listeners, I am not behind it. I am out of there. Gone. My masters are the original."
MQA is at least in some situations simply an upsampled version with a licensing fee slapped on.....

This is a misdirection at best, someone at Warner ok'd having Neil's music converted to MQA without his knowledge. The files were 44.1khz MQA (which I agree seems rather pointless, as why bother. I suspect its due to blanket licence agreements between Warner/Tidal). There's plain old MQA and there is MQA Studio masters, these were plain old MQA. I agree Tidal's choice, to use the label 'Master' can be misleading, I'm going to give them the benefit of the doubt that they mean "the best copy they have" rather then the actual "recording master"

MQA incurs an additional cost to you. You are paying for the licensing fees that are tacked on to products to get MQA support, and at every other step in the process. A good post from the manufacturer Linn is available here: https://www.linn.co.uk/blog/mqa-is-bad-for-music Given as we have now demonstrated that MQA is NOT a substitute for native HiRes content, its hard to argue that MQA is doing much more than charging you for a sub-par version of something you already had (native hires music). If you want the best quality, demand native hires releases, not licensed, closed-source, proprietary compression. Schiit audio has also spoken on it: https://www.schiit.com/news/news/why-we-wont-be-supporting-mqa

I agree that paying license fees is a pain, but what are these fee's really? $99 DAC's have MQA, so it can't really be that much. MP3 incurred a license cost until a few years ago. Licensing fee's are a way for innovators to make a business. Sure it could be open source, though it would be much harder for them to survive as a business and like MP3 it may well become that way eventually.

MQA is actually probably worse than native playback. MQA makes it basically impossible to obtain a "normal" and MQA version of the same hires file. BUT, Stereophile did manage to convince them to send an MQA encoded single-impulse file. Their testing showed three things:
1 - Playing back an MQA encoded file on a non-MQA dac caused issues, and created an asymmetric impulse response.
2 - Playing it back on an MQA capable dac, it was minimum phase, not linear.
3 - Playing back a NORMAL, non-MQA encoded impulse response file, with the MQA filter turned on on the DAC, produced an IDENTICAL result to the MQA file, suggesting that MQA is nothing more than a basic minimum-phase upsampling filter in this situation, and absolutely nothing to do with the source file. https://www.stereophile.com/content/mqa-tested-part-1
There is significant evidence from multiple third party sources to show that MQA has all sorts of problems. http://archimago.blogspot.com/2018/02/musingsmeasurements-on-blurring-and-why.html

This is conjecture. "normal" and MQA files exist, go to 2L's catalogue for freely available samples. One DAC had its MQA filter on all the time, all (or all I've seen) don't do this anymore. This is no different from you leaving your DAC in the same filter mode for all tracks. MQA attempts (if they do it right, I couldn't say) to pick the best filter for a track, and one assumes in the Mytek implementation it defaults to one if it doesn't have that information. Whether linear phase or minimum phase is really better is a whole different discussion.

As to the original "analysis" comparing the bits between a plain FLAC / MQA file makes no sense. One should really compare the analogue output, but even then this isn't conclusive. Do we care what noise is left over outside of the audible band? AFAIK the USP of MQA is not providing audible information above 20khz, but to provide the timing information of higher res content. Its meaningless to diff the bits between a file that is downsamples to 44.1 and an MQA file (folded or unfolded), of course they are going to have different bits, the bit that is important is do they actually sound different? Are there any proper controlled tests that can confirm anyone can tell them apart?

I would much prefer MQA was an open source solution, that there's wasn't a premium to use it. 99% of the software I use is open source, but there is room for the 1%, and maybe MQA fits in there.
 
Last edited:
OP
GoldenOne

GoldenOne

Active Member
Joined
Jun 25, 2019
Messages
201
Likes
1,394
I don't believe MQA is manna from heaven, or the answer to audiophile dreams. More likely it's a way for the record industry to have tighter controls over the release of hi res music, while still holding on to the originals. Fine.

But the attempts at "proof" showing it to be inferior are poorly constructed and would be torn apart in any other area, if it wasn't for the fact that everyone loves to hate MQA.

In reverse order



Did any new format have to prove how it worked to the general public, maybe I missed that. People still pay a premium for DSD files, which are proven (its a fact) to be upsampled from the DXD (PCM) masters and we don't read about how its a scam every few weeks. I couldn't say what the motivation for restricting digital out is, but one could easily record the analogue out with an ADC and compare to a known like for like source such as 2L. Plenty of folk with such devices around, yet I've not seen this done?



This is a misdirection at best, someone at Warner ok'd having Neil's music converted to MQA without his knowledge. The files were 44.1khz MQA (which I agree seems rather pointless, as why bother. I suspect its due to blanket licence agreements between Warner/Tidal). There's plain old MQA and there is MQA Studio masters, these were plain old MQA. I agree Tidal's choice, to use the label 'Master' can be misleading, I'm going to give them the benefit of the doubt that they mean "the best copy they have" rather then the actual "recording master"



I agree that paying license fees is a pain, but what are these fee's really? $99 DAC's have MQA, so it can't really be that much. MP3 incurred a license cost until a few years ago. Licensing fee's are a way for innovators to make a business. Sure it could be open source, though it would be much harder for them to survive as a business and like MP3 it may well become that way eventually.



This is conjecture. "normal" and MQA files exist, go to 2L's catalogue for freely available samples. One DAC had its MQA filter on all the time, all (or all I've seen) don't do this anymore. This is no different from you leaving your DAC in the same filter mode for all tracks. MQA attempts (if they do it right, I couldn't say) to pick the best filter for a track, and one assumes in the Mytek implementation it defaults to one if it doesn't have that information. Whether linear phase or minimum phase is really better is a whole different discussion.

As to the original "analysis" comparing the bits between a plain FLAC / MQA file makes no sense. One should really compare the analogue output, but even then this isn't conclusive. Do we care what noise is left over outside of the audible band? AFAIK the USP of MQA is not providing audible information above 20khz, but to provide the timing information of higher res content. Its meaningless to diff the bits between a file that is downsamples to 44.1 and an MQA file (folded or unfolded), of course they are going to have different bits, the bit that is important is do they actually sound different? Are there any proper controlled tests that can confirm anyone can tell them apart?

I would much prefer MQA was an open source solution, that there's wasn't a premium to use it. 99% of the software I use is open source, but there is room for the 1%, and maybe MQA fits in there.
In regards to the DSD Comparison, that's quite a different topic.

DSD is a completely different method for digital audio and isn't pcm at all.
The benefit of DSD on a 1-bit capable dac is that you effectively 'bypass' any oversampling and delta sigma modulation being done by your dac as the signal is already PDM.

This of course means you're reliant on the quality of the file. If it was upsampled from a high sample rate source, using advanced noise shapers and modulators that could not be run on a dac in real-time (such as those available in tools from Merging or Signalyst) then it's quite likely the result would be better than letting your dac convert the DXD itself.
If it was done with more basic tools then possibly not and indeed wouldn't be worth using.
And how audible the difference is would be a different question.

But in any case the point is DSD is open source, doesn't to my knowledge require licensing fees to use it, and the differences in mechanics of how its working and why it can be good/bad are clear and we can discuss the merits of the format openly and with plentiful resources and tools.

With mqa that simply isn't the case. It's working on the basis of a "trust me bro it sounds better". And we have no access to any full encoding or decoding tools ourselves.

There is no information available as to what it is doing, all the various testing done so far shows various issues and while inconclusive the unfortunate truth is that without mqa making available more information other than marketing fluff its a bit tricky to know WHAT to test for other than if its different from native hires content, which it clearly is as demonstrated by various people.


In regards to the >20khz noise I partially agree. No its likely not all that important. But the fact that it's there is still concerning.
But the fact that there are notable differences throughout the audible band especially that weird 11.5-13.5khz band of wonkiness, is the main issue
 

nimar

Active Member
Forum Donor
Joined
Jan 25, 2021
Messages
213
Likes
199
Location
Ontario, Canada
In regards to the DSD Comparison, that's quite a different topic.

DSD is a completely different method for digital audio and isn't pcm at all.
The benefit of DSD on a 1-bit capable dac is that you effectively 'bypass' any oversampling and delta sigma modulation being done by your dac as the signal is already PDM.

This of course means you're reliant on the quality of the file. If it was upsampled from a high sample rate source, using advanced noise shapers and modulators that could not be run on a dac in real-time (such as those available in tools from Merging or Signalyst) then it's quite likely the result would be better than letting your dac convert the DXD itself.
If it was done with more basic tools then possibly not and indeed wouldn't be worth using.
And how audible the difference is would be a different question.

But in any case the point is DSD is open source, doesn't to my knowledge require licensing fees to use it, and the differences in mechanics of how its working and why it can be good/bad are clear and we can discuss the merits of the format openly and with plentiful resources and tools.

With mqa that simply isn't the case. It's working on the basis of a "trust me bro it sounds better". And we have no access to any full encoding or decoding tools ourselves.

There is no information available as to what it is doing, all the various testing done so far shows various issues and while inconclusive the unfortunate truth is that without mqa making available more information other than marketing fluff its a bit tricky to know WHAT to test for other than if its different from native hires content, which it clearly is as demonstrated by various people.


In regards to the >20khz noise I partially agree. No its likely not all that important. But the fact that it's there is still concerning.
But the fact that there are notable differences throughout the audible band especially that weird 11.5-13.5khz band of wonkiness, is the main issue

Fair point re DSD being open source, though I still don't believe it serves a purpose in playback, its intended as a recording medium

Not saying MQA aren't a bit shifty, what company that is trying to sell something isn't.

The diff between 11-13khz is interesting, but still not conclusive. You don't know which is "correct", just that there is a diff, what if you 44.1 file was downsamples with an algorithm that caused this? Or the fact that the MQA file started with a different bitrate master?

As for software that can do the unfolding:

https://code.videolan.org/mansr/mqa/

Although last time I looked, only the first unfold worked correctly, the second unfold produced garbage.

My predicament is that in Canada there are no alternative hi-res streaming services so I've got to learn to deal with MQA for now. My hope is that eventually they go out of business and it becomes public domain and or eventually the'll be a more fleshed out version of mansr's code and we can all just use MQA without a special dac.
 
Last edited:
Status
Not open for further replies.
Top Bottom