MQA technical analysis

[jabbr]

9 minutes ago, mansr said:

We're talking about the interpolation filter used by the "renderer" when upsampling from 96 kHz to whatever higher rate it feeds the DAC chip.

 

So assume a room correction kernel. (End to end correction as includes playback chain) ... these exist and are / can be applied during upsampling? 

 

Something else? I am sure the details are different — trying to get at big picture differences, prior art etc.

 

Bottom line: if we are really talking end-to-end, you want the deconvolution to include the very important aspects of speakers,room etc which are far far more important than DAC to DAC differences — and DACs already have many scores of filters to choose from.

[PeterSt]

 

34 minutes ago, mansr said:

The DAC then applies what is allegedly its version of the indicated filter number on playback. From the official explanations, one is led to believe that the actual filters applied by the DAC are tweaked such that a given filter number provides the same analogue output response on every DAC (the "end-to-end" spiel). The measurements and firmware dumps we've done suggest that there is in fact no such per-device tuning.

 

This sort of guessing happens when we're not told the how and what.

Let's say that my info (ok, data) tells me that what happens in a last stage of "MQA licensing" is that the DAC where the license is given for, is made suit to MQA (this is 99,99% not guessing). Next it is my own assumption that the found 32 filters exist for that reason. For me it is a logical combination of matters (so this is 99,99% guessing but say 50% plausible). 

 

That this thus NOT suit dedicated filters or whatever people may have in mind "per track", only follows from logic for me. That you thus also can not see the reality in "32 filters for all" (tracks) is again logic hence probably correct.

However, What keeps on lacking in the stories is the "deblurring", which (for me) most obviously won't happen by any of these 32 filters - the contrary (they are not counteracting filters for track data). So the "deblurring" has to happen somewhere else, and I say that it already has been done in the file (track) which is handed to us (this follows from deduction and at this moment I 100% think this is so *assumed* it happens in the first place anywhere (!)).

 

If we combine all, then the steeper transient etc. blahblah MQA files handed to us, must counteract the DAC's blurry filtering (like any large tap filter so-called "blurs") and it is done by a. upsampling with b. one of the 32 filters. That this implies (huge) aliasing is just a decision (or Meridian hobby horse).

 

Do I make sense ?

[mansr]

11 minutes ago, jabbr said:

So assume a room correction kernel. (End to end correction as includes playback chain) ... these exist and are / can be applied during upsampling? 

 

Something else? I am sure the details are different — trying to get at big picture differences, prior art etc.

 

Bottom line: if we are really talking end-to-end, you want the deconvolution to include the very important aspects of speakers,room etc which are far far more important than DAC to DAC differences — and DACs already have many scores of filters to choose from.

Speaker/room correction isn't possible with MQA, and that's one of its major faults.

[jabbr]

2 hours ago, mansr said:

Speaker/room correction isn't possible with MQA, and that's one of its major faults.

Yes, and I am saying that the appropriate place to apply this correction is at upsampling. One can apply upsampling, correction(s), digital crossovers etc at one point.  That would be real end-to-end.

[FredericV]

26 minutes ago, jabbr said:

Yes, and I am saying that the appropriate place to apply this correction is at upsampling. One can apply upsampling, correction(s), digital crossovers etc at one point.  That would be real end-to-end.

That's what I do. Upsample to a very high resolution my DAC still supports (24/352.8) and run one simple parametric equi filter on the outcome to kill the room mode which in my case is 33 Hz being boosted with almost 10dB by the room. So this correction is done in the digital domain before sending the PCM to the Metrum DAC.

This is a good test track that will hit your room modes easily:
 

So to make this 33 Hz problem go away, I could put a lot of bass traps in the room, or just one line of sox filter recipes.

All measurement systems that I used show the exact problem: Velodyne's Digital Drive measurement system, REW, Anthem ARC, Dirac Live: 33 Hz is a big problem in my room. This is why for a a long time I could get away with the very small Marten Duke 2 monitors, which were rolling off somewhere below 40 Hz, but the room mode compensated for it.

With the Kryptons which go all the way down to 22 Hz, I can't get away with it, and not being allowed to correct this room mode, a right which MQA takes away: no thanks.

I use Amphion Krypton 3 on a Vitus SS-025 power amp connected to Metrum's flagship Adagio DAC:

 

 

[mansr]

7 minutes ago, jabbr said:

Yes, and I am saying that the appropriate place to apply this correction is at upsampling. One can apply upsampling, correction(s), digital crossovers etc at one point.  That would be real end-to-end.

Linear filters can be arbitrarily ordered or merged without affecting the end result.

[PeterSt]

 

On 2-11-2017 at 7:46 PM, mansr said:

On 2-11-2017 at 7:38 PM, GUTB said:

Just did another MQA vs non-MQA comparison through Roon with a DFR+Jitterbug to a TH900.

 

https://tidal.com/album/78913887

 

If you can't hear the MQA version is better I don't know what to tell you.

Is it the same master? Many of the MQA versions are made from better masters, which is why they sound better.

 

I did not listen but got intrigued by the fact that the MQA version compresses 3dB more than the normal Redbook; both taken from Tidal and both have the same track lengths to the 1/100 of a second. This is the normal Redbook : https://listen.tidal.com/album/78899595.

 

This album unveils a novelty to me : the compression ratios are different per track. And let's say that I never expected such a thing (never investigated it either).

With the remark that this will be opera and will be more dynamical because of it (average SPL is relatively low but close to full digital headroom is utiliized), a few tracks carry the same compression, some differ 3 dB, more differ 5dB and some even differ 10dB.

because the two longer tracks in there (8+ and 9+ minutes) coincidentally compress the same, the average that the MQA compresses more is only 3dB.

 

Since the "functional" masters as such won't differ (I am as far as being sure about that by now), the process of creating the MQA version (the "master" we receive) must be dynamical to the content it processes. Otherwise this wouldn't differ (so much) per track.

 

The difference is so large in some tracks that I can not imagine the MQA sounding better (this was the claim of @GUTBin that thread) and on some other hand the difference in SQ should somehow be huge. So although I am not into opera at all, I am curious how these compare.

 

 

On 2-11-2017 at 7:38 PM, GUTB said:

Just did another MQA vs non-MQA comparison through Roon with a DFR+Jitterbug to a TH900.

 

https://tidal.com/album/78913887

 

If you can't hear the MQA version is better I don't know what to tell you.

 

[Abtr]

1 hour ago, PeterSt said:

 

 

I did not listen but got intrigued by the fact that the MQA version compresses 3dB more than the normal Redbook; both taken from Tidal and both have the same track lengths to the 1/100 of a second. This is the normal Redbook : https://listen.tidal.com/album/78899595.

 

This album unveils a novelty to me : the compression ratios are different per track. And let's say that I never expected such a thing (never investigated it either).

With the remark that this will be opera and will be more dynamical because of it (average SPL is relatively low but close to full digital headroom is utiliized), a few tracks carry the same compression, some differ 3 dB, more differ 5dB and some even differ 10dB.

because the two longer tracks in there (8+ and 9+ minutes) coincidentally compress the same, the average that the MQA compresses more is only 3dB.

 

Since the "functional" masters as such won't differ (I am as far as being sure about that by now), the process of creating the MQA version (the "master" we receive) must be dynamical to the content it processes. Otherwise this wouldn't differ (so much) per track.

 

The difference is so large in some tracks that I can not imagine the MQA sounding better (this was the claim of @GUTBin that thread) and on some other hand the difference in SQ should somehow be huge. So although I am not into opera at all, I am curious how these compare.

 

 

 

 

Hi Peter. Like @mansr said, nearly all MQA files on Tidal are created from compressed so called remastered versions of the original. To some this may sound better. To me, in my system, it almost never does..

[Jud]

3 hours ago, PeterSt said:

a few tracks carry the same compression, some differ 3 dB, more differ 5dB and some even differ 10dB.

 

So these MQA tracks may be louder during softer passages.

[PeterSt]

5 hours ago, Abtr said:

Hi Peter. Like @mansr said, nearly all MQA files on Tidal are created from compressed so called remastered versions of the original.

 

No Sir, they are not. The masters are often the same, but what we receive from them are processed by MQA. However, *if* other masters as such have been used for the MQA base (for processing) they are the better ones. At least this is my own consistent finding. Never they are worse, but with this one possibly as the exception. That's why I jumped on it.

 

Btw, I tried listening to it, but it seriously is nothing for me, this opera. Maybe an other day.

 

4 hours ago, Jud said:

So these MQA tracks may be louder during softer passages.

 

Yes, that should be the case.

 

Thanks,

Peter

[Abtr]

5 hours ago, PeterSt said:

No Sir, they are not. The masters are often the same, but what we receive from them are processed by MQA. However, *if* other masters as such have been used for the MQA base (for processing) they are the better ones. At least this is my own consistent finding. Never they are worse ...

 

Hm.. I wish that was true. The dynamic range of so called 'remastered' albums is almost always compressed relative to the original version. The labels push such remastered versions to the point that the original albums cannot be bought new anymore and often are not available on e.g. Tidal (one reason to hold on to your old CDs). So it wouldn't make sense if the labels use uncompressed originals as the basis for MQA-ed versions if remastered versions are available.

 

I will do some more MQA listening, but in my admittedly limited experience with MQA, if a remastered version is available, then the MQA version sounds like the compressed remastered version and the volume must be set lower because it is louder than the original; not exactly like it sounded in the studio..

[PeterSt]

4 hours ago, Abtr said:

Hm.. I wish that was true. The dynamic range of so called 'remastered' albums is almost always compressed relative to the original version.

 

I fully agree with that. Now with that as a base (do not forget it), try to hear what I am telling (you) :

 

a. The "remasters" MQA uses (*if* that is the case) are the BETTER ones, not the worse ones. A sort of : back to basics. This is a good thing because now we have a better master than we were used to. It is reason alone for the MQA version to sound better.

(read again what mansr said because he says the same)

 

b. I found that in many cases (which may end up to be all cases), no real remaster as such has been used. Instead it is a processed version of one of the versions hanging around on Tidal (which in the end is one of the versions hanging around at the owning labels).

Thus, we take a version, push it through some manipulating processor, and what comes out is the version we receive. This is not a remaster; it is a processed version of an existing master and maybe we should forbid that this exists.

 

c. In either case MQA processes the file/album. So what we see is an MQA'd version and we should not call it a master of any kind. It is the opposite because it has been processed - even in lossy fashion.

 

d. Another master as such would be some mastering (up to mixing !) engineer who takes all the raw data and processes it such that it will be more bassy or whatever to his skills and liking. Usually this is not really remixing, but often the separate tracks (like for the bass guitarist) are taken and blended in differently from before.

This is NOT what MQA is doing at all, as it is impossible for the amount of work. So MQA utilizes one of the other existing masters and to my finding the best = least compressed ones. This includes ones which were never produced on to CD.

 

e. That album of the subject is the rare occasion where I see that the MQA version as delivered to us, is more compressed than the original, of which I can find one version only : on Tidal again. They are by (my) guarantee from the very same master, which is proven by the track lengths. And *because* it is the same master, it is odd to notice that MQA makes it more compressed. But why not, it processes the file(s) one way or the other.

 

New to this thread is that this works out for the better as the original one is too dynamical to my finding. Therefore it shouts or the exhaling voice is too loud for the level it requires (Crime of the Century is also unplayable for that and thinking of it, here too MQA delivers a way more compressed version - this album in native version though, does not utilize even half of the digital headroom, so it is just wrong).

 

4 hours ago, Abtr said:

if a remastered version is available, then the MQA version sounds like the compressed remastered version

 

Explicitly not my finding, with the notice that this is my preselection (and I always have real time on screen what the compression ratio is of any album I play).

More compressed albums than the ones I am used to, I don't even play (I hear it right away as well).

So what seems to be wrong with your statement is that it should say : when a least compressed version is available, MQA uses that. Even if it never saw daylight (OK, that I could find). But there is more going on, and the persons who select the "base" masters for MQAs processing, must be intelligently looking at what version to use. So if it would be "just the least compressed" then they should have taken the original Crime Of the Century. But they didn't.

[JensPH]

Am I the only one genuinely confused here? Several of the TIDAL MQA Masters sound not just different but clearly better on my gear and so much so, that family members who were oblivious to the fact that I only just started a TIDAL 30 day trial asked what was playing.  I have 24-bit versions of everything available but even they are often squashed with compressors and limiters but the TIDAL version is much clearer and "airier" or more three dimensional. Those recordings which are only available in Red Book CD quality and sound bad both as CD and download sound exactly the same via TIDAL HiFi.  

I have no trust in the MQA mumbo jumbo as the end resolution clearly is way below 24 bit, but something is different since it is possible to hear the difference from the kitchen and my guess is that the masters are way less limited or for some reason have far less clipping but it also sounds like there could be used a slightly different EQ or mix.

 

Does anyone have any proof of what is actually going on or has anyone been able to put a decoded TIDAL stream through the Crest Value/Dynamic Range test? 

[mansr]

Many of the MQA albums are from different masters than their non-MQA counterparts. I did a detailed analysis of one track here:

Among other differences, the MQA version is 8 seconds longer.

[labjr]

18 minutes ago, mansr said:

Many of the MQA albums are from different masters than their non-MQA counterparts. I did a detailed analysis of one track here:

Among other differences, the MQA version is 8 seconds longer.

I can't believe the record industry and publications just went along with this  scam that appears to have such lack of transparency, both literally and figuratively!

[crenca]

48 minutes ago, JensPH said:

I have no trust in the MQA mumbo jumbo as the end resolution clearly is way below 24 bit, but something is different

 

 

 

What's different is the mastering.  Even a marginally improved master will probably sound better at 320 bit mp3 than the marginally worse master at 24/192 PCM.

 

When comparing an apple-to-apple same master, the difference between 16/44, 24/96, and MQA is very difficult to flush out.  Bob Jame's "The New Cool" is one such album on Tidal.  I have the high res, and Tidal streams both the 16/44 and MQA.  It is a recent album and while I can't "prove" it, I believe it is clear that the same master is used across all three encodings.  Might not be your cup of tea (its jazz), but listen to it.  If you can hear large significant differences, then it is all between the ears  

[PeterSt]

5 hours ago, mansr said:

Many of the MQA albums are from different masters than their non-MQA counterparts. I did a detailed analysis of one track here:

Among other differences, the MQA version is 8 seconds longer.

 

Yes, but that was not the answer. The answer is a few posts further in there (my post). We must compare apples with apples.

 

Anyway, the moral is that the masters indeed are often different. But as often they are not, but processed by MQA (say DSP) before they get to us.

[JensPH]

13 hours ago, PeterSt said:

Anyway, the moral is that the masters indeed are often different. But as often they are not, but processed by MQA (say DSP) before they get to us.

 

From what I've read of the documentation on MQA I can't figure out why the processing can make the differences I hear. I tend to think, that MQA probably requires a less clipped master than the loudness ruined ones used by Apple, Spotify and in CD production. Maybe like SACD/DSD which can't handle clipped material - Depeche Mode "Ultra" is a perfect example of that. The SACD is the only version worth listening to apart from the first edition vinyl. 

[PeterSt]

8 hours ago, JensPH said:

I tend to think, that MQA probably requires a less clipped master than the loudness ruined ones

 

Hmm ... You can well be right there but with no reason for it other than I have been thinking the very same myself . This thinking was quite explicit, but by now I forgot the how or why. Something with expansion (like emphasizing high frequencies and the digital room required for that).

 

[Don Hills]

On 1/23/2017 at 5:20 AM, mansr said:

I found something interesting. The decoder has the provision to swap bits [4:11] and [12:19] in the output and signal to the renderer to reverse this shuffle. If this mode is triggered, only the top 4 bits are passed intact, so while the music would be recognisable, it would sound awful. I don't know what might enable this as I have not seen it with any of the samples I've tested (hardly surprising). Explain how this isn't DRM.

 

@mansr, did you ever get any further with this? I originally read it that the bit reversal was present in the undecoded stream, but on re-reading you appear to be saying that the core decoder reverses the bits before feeding them to the renderer. This would appear to be of limited utility, it simply enforces the use of an MQA renderer. It also sounds trivial to work around.

[mansr]

11 minutes ago, Don Hills said:

@mansr, did you ever get any further with this? I originally read it that the bit reversal was present in the undecoded stream, but on re-reading you appear to be saying that the core decoder reverses the bits before feeding them to the renderer. This would appear to be of limited utility, it simply enforces the use of an MQA renderer. It also sounds trivial to work around.

Having never seen this feature used with an actual MQA file, I can't say what the undecoded stream would look like. I only know about it because the renderer contains code to undo this bit shuffling. Moreover, newer version of the Bluesound firmware seem to have dropped this feature, and it is also not present in the Dragonfly. Maybe they had initially planned to allow software decoding in order to support low-powered render-only devices like the Dragonfly while still requiring some kind of licensed hardware. That would certainly be an attractive proposition to anyone of the money-grubbing persuasion.

[Don Hills]

Thank you.

I am now confused at a higher level, because in the vaporware thread I recall you saying that  the encryption could be applied to "higher" bits of the undecoded (streamed) data. I'll need to review the threads and see what I missed.

I'm trying to work out the place of the encryption in the process. My current understanding is as follows. Can you answer a true/false for each step please?

 

(1) Currently, the "high res" part (> 24 KHz) is encoded into 8 bits and then

(2) encrypted before being added below the upper 15 bits of each sample. (Bit 8 reserved for the control stream.)

(3) The encryption can be applied to more of the 24 bits of the sample, effectively reducing the bit depth of the 0-24 KHz data. For example, there might be 7 bits of 0-24 KHz audio and 16 bits of encrypted data. This would require the control stream to be positioned "higher" in the 24 bit word, above the encrypted portion, and the format does apparently allow for this. 

 

Thanking you in anticipation...

[mansr]

5 minutes ago, Don Hills said:

I am now confused at a higher level, because in the vaporware thread I recall you saying that  the encryption could be applied to "higher" bits of the undecoded (streamed) data. I'll need to review the threads and see what I missed.

I'm trying to work out the place of the encryption in the process. My current understanding is as follows. Can you answer a true/false for each step please?

 

(1) Currently, the "high res" part (> 24 KHz) is encoded into 8 bits and then

8 bits or more. I have seen files where 9 bits are used in addition to the control stream. The format allows for even more.

 

5 minutes ago, Don Hills said:

(2) encrypted before being added below the upper 15 bits of each sample. (Bit 8 reserved for the control stream.)

I don't know how this data stream is encoded. The final step before insertion into the PCM stream is to xor the bits with a simple pseudo-random sequence. There may or may not be actual encryption prior to this. It doesn't really matter, though. Since the format is secret, the decoder has to be reverse engineered either way.

 

5 minutes ago, Don Hills said:

(3) The encryption can be applied to more of the 24 bits of the sample, effectively reducing the bit depth of the 0-24 KHz data. For example, there might be 7 bits of 0-24 KHz audio and 16 bits of encrypted data. This would require the control stream to be positioned "higher" in the 24 bit word, above the encrypted portion, and the format does apparently allow for this.

There is a mechanism for encrypting part of the PCM data, i.e. the top 15 bits, using the Salsa20 cipher with the control stream providing the decryption key (yeah, very secure). If this key is present, the decoder performs decryption. This is independent of whatever encoding is used for the compressed high-frequency data in the low 8 bits. Authentication is also a separate process.

[Shadders]

Hi,

Latest copy of Hifi News states that : "Aliasing does occur, but above 48kHz"

 

I was under the impression that aliasing occurred from 24kHz to 48kHz ?

 

The article is on MQA on CD.

 

Thanks and regards,

Shadders.

[mansr]

12 minutes ago, Shadders said:

Latest copy of Hifi News states that : "Aliasing does occur, but above 48kHz"

What a ridiculous statement. Since the sample rate is 96 kHz, nothing can occur above 48 kHz. Perhaps they meant imaging (in the upsampling). That is indeed above 48 kHz.