TIDAL Brings MQA, Matt analysis

[Matthew J Poes]

You can get the benefits of MQA through the phone if you use the USB camera kit and a 24 bit/96khz dac.

I’ve been analyzing the streams from Tidal and Qobuz. I can say this much, if the music is of a true Hd source, the MQA Tidal Master has the same basic spectral balance and dynamic range as the original flac file from Qobuz. Almost nothing I’ve analyzed has much ultrasonic content, but I’ve found a different benefit to be evident. The HD tracks have a smoother and I assume more accurate spectral balance between 10 and 20 kHz. The 44.1khz tracks have an excess of energy in that range. A digital signals expert has been helping me analyze the data and we’ve drawn the same conclusion. There doesn’t seem to be anything musical above 20khz but the HD tracks are more accurate below 20khz. The advantage of the HD tracks, if the content is really HD is more accurate HF’s in the audible range. THe MQA streams from Tidal appear basically identical to that of the Qobuz tracks, and both look different from that of their red book cd quality counterparts.

I’ll share some analysis graphs later to show.

 

[Matthew J Poes]

Qobuz Flac of John Williams Jurassic Park 24/96


Same song, Tidal Master 24/96


Tidal HiFi 16/44.1


The approach I used to do this is a quicker way and restricts it to 16 bit, but that has no impact on anything meaningful. The musical content here does not exceed the dynamic range of dithered 16 bit audio. That and I was only interested in showing you the HF content differences. The spectral content isn't obviously different between the MQA and Flac files, but clearly shows a difference from the 44.1khz file, which has no spectral content above that point. I think it's also fair to say there is very little musical content past 6-10khz, there is something there, but it really looks mostly like the harmonics. Most of the musical energy is actually down low. Whatever is up there, the MQA and Qobuz HD files have a bit less energy between 15 and 21khz, and when you extract the difference from the files in that range, you see it's excess noise (i.e. it's not musical). The only thing I've found evidence of in the MQA tracks vs FLAC (and I'm not confident that I have anything) is de-emphasis between 2-3khz. I need to analyze a lot more tracks before I feel confident it isn't a fluke.

In any case, I shared this just to make the point that my analysis shows you shouldn't necessarily worry to much about MQA. As far as HD content goes, with real content, I'm not seeing anything material to be concerned about so far. There isn't a lot of music up there, but it's at least not different from Qobuz Flac.

 

[ddude003]

I was wondering if you could elaborate on this statement from above...

" There doesn’t seem to be anything musical above 20khz but the HD tracks are more accurate below 20khz."

Are you talking about the sample equipment and music sources (HD, MQA and Qobuz) you are testing or is this a statement about music in general? Many musical instruments (strings, woodwinds, brass and percussion) produce energy to 40 kHz and/or above... And what effects do ultrasonics have on the human body and mind... Is this the realm of "air" and "sparkle" and "a sense of venue" as well as brain state changes?

 

[Matthew J Poes]

I was wondering if you could elaborate on this statement from above...

" There doesn’t seem to be anything musical above 20khz but the HD tracks are more accurate below 20khz."

Are you talking about the sample equipment and music sources (HD, MQA and Qobuz) you are testing or is this a statement about music in general? Many musical instruments (strings, woodwinds, brass and percussion) produce energy to 40 kHz and/or above... And what effects do ultrasonics have on the human body and mind... Is this the realm of "air" and "sparkle" and "a sense of venue" as well as brain state changes?

In the analysis of the file, we took the files and created differencefiles in Matlab. We could then examine those difference files against the musical content itself in the full files. For example, do they match up to the harmonics of instruments. What we found was it's just noise. It isn't musical content. This isn't about instruments having ultrasonics, this is about these particular tracks that I analyzed.

I was also talking specifically about the HF energy I found that was different between the 44.1khz file I analyzed and the 96khz file version. In this particular track, the difference was actually just noise.

I've actually analyzed quite a few files now and a lot of them show evidence of analogue bandwidth limiting. I would need the help of more experts with knowledge of the bandwithof the tape equipment, digital recorders, microphones, preamps, etc. to figure out what I am seeing, but I've found a bunch of music that looks as if it was recorded with equipment that itself rolls off above 12khz or so. I've also found a bunch that show a notch around 22khz that is fairly wide and shows evidence of aliasing above that (a mirror image). I have a feeling it has been upsampled. As such, the recording itself doesn't have any actual musical content above 22khz.

Here is another comparison graphic, but....I haven't had a chance to level match the files yet. Until I can do that, you need to ignore the absolute level differences and only look at the relative level differences. For whatever reason, the 44.1khz file from Qobuz is at a lower level above 1khz than that of the other files.



What I think is most interesting is that the HD files are far smoother in the HF spectral response than the CD quality file is. On top of that, it drops like a brick at around 19khz, even though its a 44.1khz file. The HD files show continued content that smoothly extends out to that spike and then keeps going. That spike? No idea what it is, but lots of guesses. It shows up in some tracks but not others. It isn't my system as I've confirmed it from different playback sources and differnt recording computers and methods of recording the tracks. In this case, my best guess is that its a switching frequency associated with some of the recording equipment.

@ddude003 please don't misinterpret what I'm saying about HD music here. I am defending it's existance. I'm making no statements or passing no judgement on if the ultrasonics are audible or reproducable in theory. What I'm saying is that in analyzing the files, I found that the CD quality tracks had what I and a Digital Signals expert think is a sign of less accurate sound between 10 and 20khz or so. That we also found evidence that the recordings themselves are not really of HD quality, lacking much in the way of ultrasonic content. What I'm showing you here is litterally the best thing I've found on Qobuz and Tidal so far, everything else has been MUCH worse.

 

[Matthew J Poes]

I just realized this is really confusing without seeing better what I'm talking about.

Ok so....It isn't the ultrasonics that I am saying is just noise. In the CD quality tracks, which contain no ultrasonics, there is an excess of energy between 10 and 20khz in many of the files we analyzed. When we examined it, it was just noise. The stuff above 20khz varied. Some of it was aliasing artifacts, some of it was noise, and some of it was musical harmonics.



This image with the little additions help illustrate the differences. I level matched the 16 bit file so you can see the area that is spiking above the 96khz files.

 

[ddude003]

Thank you Matthew for your clarification and additional thoughts on the subject of High-resolution Digital Music and comparisons between various digital music sources... Like you, I have noticed that the quality is all over the place for various reasons... I appreciate your efforts in bringing these issues to light... There is still a lot of room for innovation in Compression, Decompression, Encoding, Decoding and Upscaling of Analogue and Digital Signals...

 

[Grayson Dere]

Qobuz Flac of John Williams Jurassic Park 24/96
View attachment 12740

Same song, Tidal Master 24/96
View attachment 12742

Tidal HiFi 16/44.1
View attachment 12741

The approach I used to do this is a quicker way and restricts it to 16 bit, but that has no impact on anything meaningful. The musical content here does not exceed the dynamic range of dithered 16 bit audio. That and I was only interested in showing you the HF content differences. The spectral content isn't obviously different between the MQA and Flac files, but clearly shows a difference from the 44.1khz file, which has no spectral content above that point. I think it's also fair to say there is very little musical content past 6-10khz, there is something there, but it really looks mostly like the harmonics. Most of the musical energy is actually down low. Whatever is up there, the MQA and Qobuz HD files have a bit less energy between 15 and 21khz, and when you extract the difference from the files in that range, you see it's excess noise (i.e. it's not musical). The only thing I've found evidence of in the MQA tracks vs FLAC (and I'm not confident that I have anything) is de-emphasis between 2-3khz. I need to analyze a lot more tracks before I feel confident it isn't a fluke.

In any case, I shared this just to make the point that my analysis shows you shouldn't necessarily worry to much about MQA. As far as HD content goes, with real content, I'm not seeing anything material to be concerned about so far. There isn't a lot of music up there, but it's at least not different from Qobuz Flac.

This is really awesome how you analyzed the different files. Out of curiosity, and please forgive me if you've mentioned it already in this thread, but when you listened to the Qobuz 24/96, Tidal Master 24/96, and Tidal HiFi 16/44.1 versions which did you enjoy the most? Thanks, Matthew!

 

[Matthew J Poes]

This is really awesome how you analyzed the different files. Out of curiosity, and please forgive me if you've mentioned it already in this thread, but when you listened to the Qobuz 24/96, Tidal Master 24/96, and Tidal HiFi 16/44.1 versions which did you enjoy the most? Thanks, Matthew!

Under most circumstances and with most of the songs I listened to I couldn’t tell a big difference. It was only once I found that tribute to John Williams track that I felt I could maybe hear a difference. In that case I didn’t have a preference for Qobuz he Tidal and as you can see I didn’t measure any meaningful differences either. I thought the high resolution sounded better than the standard resolution. Having said that, I’ve since found that the level of the high frequencies above about 800hz is lower in the standard resolution files from Qobuz and Tidal (at least for that John Williams tribute song). I’m going to borrow some CD’s from the library and do more analysis of all this against the CD to see if the benefit I heard is due to some kind of EQing that is being done to make the standard definition files sound less exciting.

 

[Grayson Dere]

Under most circumstances and with most of the songs I listened to I couldn’t tell a big difference. It was only once I found that tribute to John Williams track that I felt I could maybe hear a difference. In that case I didn’t have a preference for Qobuz he Tidal and as you can see I didn’t measure any meaningful differences either. I thought the high resolution sounded better than the standard resolution. Having said that, I’ve since found that the level of the high frequencies above about 800hz is lower in the standard resolution files from Qobuz and Tidal (at least for that John Williams tribute song). I’m going to borrow some CD’s from the library and do more analysis of all this against the CD to see if the benefit I heard is due to some kind of EQing that is being done to make the standard definition files sound less exciting.

That would be interesting to find out about the possible EQing going on, if any.

 

[Matthew J Poes]

That would be interesting to find out about the possible EQing going on, if any.

It would be quite the scandal so I need to analyze a lot more tracks before drawing any conclusions.

I’ve also found evidence of Tidal MQA tracks having de-emphasis at around 2-4khz that isn’t compensated for in the unfolding process in the app. It’s my understanding that MQA uses de-emphasis but I would have expected that to be eqed back. It should be quite audible. It’s shown up in every MQA file I’ve analyzed so for, but again, I want to be absolutely sure of what I’m detecting before saying anything official. Consider this just oddities in my measurement for now.

It’s not uncommon for my investigations to go down a rabbit hole with various twists and turns of my own making. Usually with some user error explaining the most scandalous issues.

 

[ddude003]

If it is not too much bother, would you mind telling us what your signal chains look like with your Matlab investigations (are you reading music tracks/files directly onto a computer running Matlab to directly analyze or?) as well as, what is the listening environment kit/setup...

And would it be a good idea for the Mods to move some of this which does not pertain directly to TIDAL Brings MQA Support to Apple's iPhone to another thread?

 

[Matthew J Poes]

If it is not too much bother, would you mind telling us what your signal chains look like with your Matlab investigations (are you reading music tracks/files directly onto a computer running Matlab to directly analyze or?) as well as, what is the listening environment kit/setup...

And would it be a good idea for the Mods to move some of this which does not pertain directly to TIDAL Brings MQA Support to Apple's iPhone to another thread?

The output of tidal and Qobuz is sent to Audacity through a virtual patching software. It’s all digital and all software based. No conversions and totally transparent. I was originally recording everything as 32 bit floating point and 384khz but that caused severe aliasing. I then eventually had to switch to native sampling recordings to avoid the aliasing problem. Those are saved out as 24 bit or 16 bit native sampling wav files. Those are then sent to either the analysis software or imported into Matlab for analysis, depending on what I’m doing.

The other approach I used was to use the virtual patch software to go from the streaming software to the analysis software, which can analyze a live stream. In that scenario, as long as the sampling frequency is a multiple of the native sampling, it works without causing aliasing.

MQA decoding is handled by the app and as such is just the first unfold.

I could try to mimic the second unfold with Audacity I suppose but I don’t expect it to do anything meaningful. Especially given how little musical content already exists in these recordings.

I haven’t tried 2L recordings. Come to think of it, that would probably be a good thing to check out. They are much more likely to contain real ultrasonic content.

 

[Todd Anderson]

Just to make sure I'm reading this correctly... TIDAL Hi-Fi lops off information right around 20kHz... the two hi-res options have info extending up into the 40khz region?

 

[Matthew J Poes]

Just to make sure I'm reading this correctly... TIDAL Hi-Fi lops off information right around 20kHz... the two hi-res options have info extending up into the 40khz region?

Not exactly. It’s not a Tidal thing. The lopped off info is the CD quality. So it could be from a Tidal Hifi sound or it could be from Qobuz CD quality. They looked the same.

I think it’s interesting that it’s lopping off HF I formation well below the nyquist frequency.

 

[Todd Anderson]

Right...

But that horizontal axis is frequency, yes?

I guess this all begs the question: does the added information above 20kHz mean anything to the human ear? It certainly means something in the bass region!

 

[Matthew J Poes]

Right...

But that horizontal axis is frequency, yes?

I guess this all begs the question: does the added information above 20Hz mean anything to the human ear? It certainly means something in the bass region!

It’s khz, so above 20khz. I think you knew that but just in case others missed that point.

So I would say don’t look at the area above 20khz. Look at the similarity at and below 20khz. Notice the differences? We don’t know with certainty which is the right one, I would need the original master to know that. What we can guess is that the smoother one is probably the more accurate one. And in this case, the HD files have a smoother HF spectrum than the redbook ones. There seems to be some evidence of excess energy in particular audible frequencies that shouldn’t be there. My guess is that a phase analysis would show the problem to be phase problems caused by the lower sampling frequency.

I mean going back to the notion of the value in ultrasonics, there is research that supports that.
https://www.researchgate.net/public...f_High_Resolution_Audio_Perceptual_Evaluation
This was the Reiss paper that found that across all comparable high quality studies, there was a small but significant effect in favor of its audibility. There is also good theoretical reason that goes back to the audible range. Control theory says you need the sampling rate to be 4 times the maximum usable frequency, not 2 times. Yet with audio we have stuck with Nyquist and 2 times.

However, if we forget all that for a moment and simply go with the practical. The vast majority of receivers only operate at a 48khz sampling rate once dsp is turned on. That includes bass management. As such they can’t reproduce ultrasonics nor take advantage of moving the filter out of the audible range. Further, my own measurements of speakers with my new mic capable of measuring ultrasonics has found that almost no speakers reproduce ultrasonics well, if at all. Many of them have breakup modes and phase issues at or just above 20khz and the response is terrible after that point.

What that means is that there are numerous bottle necks which prevents the vast majority of systems from actually reproducing ultrasonics. Even if they make a difference, it would be fair to say that the vast majority of people aren’t taking advantage of them.

My own opinion is that we need more research. First we need to better understand what exactly is audible. Is it the movement of the filter out of the audible range, the subsequently small spectral difference between 11khz and 20khz. Is it the ultrasonics themselves? If it is the latter, then you need a system dedicated to the ability to accurately reproduce ultrasonics. If that is the case, we need research into what response shape is ideal to maximize that benefit and how best to achieve it. As of right now the majority of super tweeters have very narrow dispersion and very little output much past 30khz. Most don’t even get that far. I just measured a RAAL ribbon that dropped off by 10dB just 5 degrees horizontally past 25khz. Add in air absorption at those Hf’s and I couldn’t measure meaningful ultrasonics.

 

[Todd Anderson]

Yes, I meant 20kHz.

Interesting stuff. Perhaps its just me... but I think splicing hairs becomes an exercise in futility. For me, at least, the bigger picture - and simple enjoyment of a system can easily be ruined by drilling down to the ultra minutia. That being said, a lot of what you're talking about is super interesting. Good stuff, Matt

 

[ddude003]

Matt, I wonder if some what you are seeing are what is known as 'inter-sample overs' which seem common in 44kHz CD Recordings... Who knows what happens once these are remixed, upsampled, sliced and diced for a streaming service...

I use SoX or iZotope 64-bit for my own sample rate conversion and tweaks...

Looking around for a new Separate DAC and found that a Chord Qutest performs on the bench remarkably flat from 4Hz to 46kHz... DACs in AV Receivers have a way to go to catch up... ;^)

 

[Matthew J Poes]

Matt, I wonder if some what you are seeing are what is known as 'inter-sample overs' which seem common in 44kHz CD Recordings... Who knows what happens once these are remixed, upsampled, sliced and diced for a streaming service...

I use SoX or iZotope 64-bit for my own sample rate conversion and tweaks...

Looking around for a new Separate DAC and found that a Chord Qutest performs on the bench remarkably flat from 4Hz to 46kHz... DACs in AV Receivers have a way to go to catch up... ;^)

I’m intentionally not upsampliny anything with this so it should reflect what comes directly from the software. I don’t think the software upsamples either.

I have no idea what inter-sample overs would look like on this kind of analysis to know. I could maybe look more into that and see if there is another test I can run to tell. Everything I shared here was based on a canned analysis. I have some deeper dive stuff but can’t share right now (plus it all needs to be peered still).

 

[Matthew J Poes]

Yes, I meant 20kHz.

Interesting stuff. Perhaps its just me... but I think splicing hairs becomes an exercise in futility. For me, at least, the bigger picture - and simple enjoyment of a system can easily be ruined by drilling down to the ultra minutia. That being said, a lot of what you're talking about is super interesting. Good stuff, Matt

It’s certainly a last 1% issue and I have no idea what level of enjoyment this adds. This is a shareable part of a larger project. I just thought you all would find it interesting. I obviously don’t dissect my music like this on a day to day basis.

The interesting part of this for me goes back to the research. So much that people hang their hats on around sound quality have little or no research to support the premise. This is such a controversial idea and yet it actually has pretty strong supporting evidence (that I find is frequently dismissed out of hand).

But nothing will change the fact that it is niche. Obviously high sampling rate music won’t make a sound bar, tv speakers, or Sonos speaker more enjoyable. It only applies to high end headphones or high end speaker based sound systems.

 

[ddude003]

I’m intentionally not upsampliny anything with this so it should reflect what comes directly from the software. I don’t think the software upsamples either.

I have no idea what inter-sample overs would look like on this kind of analysis to know. I could maybe look more into that and see if there is another test I can run to tell. Everything I shared here was based on a canned analysis. I have some deeper dive stuff but can’t share right now (plus it all needs to be peered still).

Matt, please don't get me wrong... I am not suggesting you are upsampling anything... I am suggesting that streaming media services and/or their outsourcing of these services may do all kinds of wild things to process and prep media for delivery...

 

[Todd Anderson]

It’s certainly a last 1% issue and I have no idea what level of enjoyment this adds.

For some small sliver of the audiophile bunch, it adds a lot. I think there's another subset that want good sound.. then beyond that it's a lump of folks that are perfectly happy listening to music on stock car radios and bluetooth speakers. I'm much more of the second crowd and can easily slip into the third.

For me, I probably start approaching #1 when it comes to movies

So what's the larger project?

 

[ddude003]

I'm not so sure I would call High-Res Audo Music Systems a niche market... Maybe a market segment... Their are a lot of players with a lot of products in this segment... And no more niche than some of the High-Def High-End TVs or Projection-based systems...

High-end things that were niche 5 years ago are common living room devices today... Is there a Moore's Law for AVSystems?

 

[Matthew J Poes]

I'm not so sure I would call High-Res Audo Music Systems a niche market... Maybe a market segment... Their are a lot of players with a lot of products in this segment... And no more niche than some of the High-Def High-End TVs or Projection-based systems...

High-end things that were niche 5 years ago are common living room devices today... Is there a Moore's Law for AVSystems?

I would consider it niche even if I’m a part of it. It’s total world wide market is $1 billion. It’s expected to remain stagnant or shrink. By comparison, home theater is about 15-20 times that size and is growing (and that is still considered a small and unhealthy market segment). Headphones alone are almost the same size market segment as home theater today.

Smart speakers is like a 10 billion dollar market and rapidly growing.

So when individual product sectors are selling ten to thirty times more product in revenue than the entire broad high end audio market, I think it’s fair to call it niche.

 

[Matthew J Poes]

For some small sliver of the audiophile bunch, it adds a lot. I think there's another subset that want good sound.. then beyond that it's a lump of folks that are perfectly happy listening to music on stock car radios and bluetooth speakers. I'm much more of the second crowd and can easily slip into the third.

For me, I probably start approaching #1 when it comes to movies

So what's the larger project?

So for you the sound quality of movies matters more than the sound quality of music?

I actually find it hard to judge absolute sound quality with movies. Music is much more where sound quality matters for me.