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

[pjug]

I think what it actually does is more like this (an illustration for 16-bit "MQA-CD" in red, solid is un-dithered area, dash is a 3-bit dither) [using Meridian's old pic]:

View attachment 133397

Thanks. In other words, any type of content in the audible band should encode fine (down to -100dB or more for MQA, less for MQA-CD, dynamic range extended by dithering similar to what redbook can do). The only signals that will throw off the encoder will be unusual ultrasonics that don't fit in the small ultrasonics box.

 

[DimitryZ]

Thanks. In other words, any type of content in the audible band should encode fine (down to -100dB or more for MQA, less for MQA-CD, dynamic range extended by dithering similar to what redbook can do). The only signals that will throw off the encoder will be unusual ultrasonics that don't fit in the small ultrasonics box.

Exactly.

Doesn't sound so scarry/suspicious all of a sudden

 

[Raindog123]

Thanks. In other words, any type of content in the audible band should encode fine (down to -100dB or more for MQA, less for MQA-CD, dynamic range extended by dithering similar to what redbook can do). The only signals that will throw off the encoder will be unusual ultrasonics that don't fit in the small ultrasonics box.

Goes without saying, my illustration is just a guesswork. Only those who implemented and tested the codec would know for sure.

(And again, a PCM waveform does not come in this frequency-domain form but rather as time-domain samples. The transformation and the result look very different, more like this coarse illustration. The 'ultrasonics' will be that fine stricture riding on top of the bigger slower 'baseband' waveform):

 

[raistlin65]

Interesting. But regardless of how it goes, we already see the eternal audiophile conundrum: "lossless in the analog domain", ok, but how then is it "much less stressful for extended listening" ?

It seems that no one (including myself ofc) can avoid somewhat contradictory opinions when it comes to audio.

Maybe because MQA says it's "better than lossless?"

https://www.mqa.co.uk/newsroom/faqs/is-mqa-lossless

So it's "less stressful" on the audiophile who is worried that lossless is not good enough. lol

Meanwhile, it can't be both "lossless" and "better than lossless." So does this mean MQA is now confirming that it's not lossless?

 

[voodooless]

One could imagine a very low compression rate audiophile lossy codec that reaches into ultrasonics and filters out the crazy noise in the DXD master to deliver exceptional sound quality. Most people would reject it out of hand.

Yes, just do a decent resample to 24/48 and be done with it.

At that point, they become practically lossless to the listener.

Indeed

 

[DimitryZ]

Goes without saying, my illustration is just a guesswork. Only those who implemented and tested the codec would know for sure.

(And again, the PCM waveform does not come in this frequency domain form but rather as time-domain samples. The transformation and the result looks very different, more like this coarse illustration. The 'ultrasonics' will be that fine stricture riding on top of the bigger slower 'baseband' waveform):

View attachment 133400

The practical limitation is ultrasonic level. And I think everyone would agree that outside of man made noise shaped files, actual natural ultrasonic content is very low.

 

[DimitryZ]

Maybe because MQA says it's "better than lossless?"

View attachment 133401

https://www.mqa.co.uk/newsroom/faqs/is-mqa-lossless

So it's "less stressful" on the audiophile who is worried that lossless is not good enough. lol

Meanwhile, it can't be both "lossless" and "better than lossless." So does this mean MQA is now confirming that it's not lossless?

Just ignore their advertising and irritation/desire to argue with it will dissipate over time.

I try to never see advertising and if forced to, to not pay attention.

 

[pozz]

When we lump MQA with lossy codecs, we commit a methodological mistake. This is because we are not differentiating the very divergent design intents of both approaches. Design intent comes first and only after that design implementation can be contextually understood. Form follows function.

Consider, that the explicit goal of a lossy codec is to, well, loose as much data as possible, while remaining audibly transparent and they operate in the baseband. The goal of MQA is to retain all the musical Information, including near ultrasonics. The LOSSY codec is actively shedding musical data, while MQA is trying to hold on to it. These are dramatically different design intents.

With such different design intents, implementation is vastly different. Perceptual codec has a complex psychoacoustic engine that looks for and discards musical detail that is judged to be masked to the listener. MQA is much simpler - it identfies the ultrasonic music limits and encodes it into the baseband LPCM, with a bit of bit-shifting. That's it (if you remove the "deblurring" step). Outside of the ultrasonic limit and noise floor, MQA makes no decisions about music or it's perception by the listener. Perceptual codec makes decisions about music audibility thousand times every second.

One could imagine a very low compression rate audiophile lossy codec that reaches into ultrasonics and filters out the crazy noise in the DXD master to deliver exceptional sound quality. Most people would reject it out of hand.

If there is a "perceptual" goal in MQA, it's not for the listener, it's to your equipment. They are attempting to hold any code/decode differences to the original at or below our systems' SNR. Judging by various tests from @Archimago and others, they appear to be succeding. At that point, they become practically lossless to the listener.

I'd disagree with the last point just based on the filters used, introduction of low level noise and various difficulties manufacturers had in getting their products to work with MQA correctly.

 

[tmtomh]

I think what it actually does is more like this (an illustration for 16-bit "MQA-CD" in red, solid is un-dithered area, dash is a 3-bit dither) [using Meridian's old pic]:

View attachment 133397

80dB of undithered dynamic range is 16dB less than 16-bit redbook.

Further, the 3 bits' worth of "dithered" range is not quite entirely dither, as the 3-bit space is necessary for MQA codec information, meaning it's not randomized. In other words, those lower three bits are compromised in a way that the 16th bit of redbook is not compromised when it's used for dither.

So the notion that this is a 100dB noise floor is questionable, misleading, or flat-out false.

 

[Raindog123]

80dB of undithered dynamic range is 16dB less than 16-bit redbook.

Further, the 3 bits' worth of "dithered" range is not quite entirely dither, as the 3-bit space is necessary for MQA codec information, meaning it's not randomized. In other words, those lower three bits are compromised in a way that the 16th bit of redbook is not compromised when it's used for dither.

So the notion that this is a 100dB noise floor is questionable, misleading, or flat-out false.

Can you sketch your levels over my cartoon? I'll pretty them up an update (if we'll all agree)

 

[pjug]

Goes without saying, my illustration is just a guesswork. Only those who implemented and tested the codec would know for sure.

(And again, the PCM waveform does not come in this frequency domain form but rather as time-domain samples. The transformation and the result looks very different, more like this coarse illustration. The 'ultrasonics' will be that fine stricture riding on top of the bigger slower 'baseband' waveform):

View attachment 133400

Isn't a simplified summary (for the first unfold) that the encoder is a band splitter, and the decoder joins the bands?
https://patents.google.com/patent/WO2017109495A1/en (Lossless bandsplitting and bandjoining using allpass filters)

 

[pjug]

For anyone who wants to take a look at the 2L-053 MQA files, @mansr generously has made them available to us. Again, the rendered files were made using some reverse engineering of a Bluesound renderer, so no ADC. (assuming I don't misunderstand)

https://drive.google.com/drive/folders/1zc5I2VJnVwBOqweOh8napqABQadnmCoe

 

[ebslo]

If you want to keep at least a pretense of fairness, then compare MQA to 24/96 FLAC, not to 24/352.

Unless I misinterpreted, did you not imply earlier that comparison even with 20/96 FLAC may be apropos?

Reposting my original question to @abdo123 below with full context, I was addressing his assertion that MQA files were smaller for equivalent perceptually relevant information, and you have answered the size part of that in the negative. The perceptual equivalence part is of course much more difficult to prove definitively as evidenced by the page count of this thread, but over the course of it I think there is enough to have reached an at least somewhat informed opinion.

It may interest you, that for the same file that Amir tested, if the original file (24 bit 88 kHz) is converted to 20 bit, keeping the sampling rate unchanged, you get essentially the same size as mqa:

45474K - 2L-145_01_stereo.mqa.flac

45195K - 2L-45_stereo_01_FLAC_88k_24b.converted.flac

71914K - 2L-45_stereo_01_FLAC_88k_24b.flac

I didn't quite grasp the ramifications at the time, and it was questioned later:

Now you have to re-encode the MQA to target 20 bits, not 24.

But if we are comparing file size vs. "musical" content (as defined by the literature on behalf of MQA), then it appears comparison between 24-bit/48kHz MQA and 20-bit/96kHz FLAC encoded PCM may indeed be valid, as the extra four low-order bits in the MQA file are being used to store extended-frequency content instead of extra sub-nyquist resolution.

 

[DimitryZ]

I'd disagree with the last point just based on the filters used, introduction of low level noise and various difficulties manufacturers had in getting their products to work with MQA correctly.

In my very low distortion, high SNR system, my amplification SINAD is in the mid 80s or so. All MQA has to do is keep any nasties at or below that and it's practically lossless to me.


And a according to @Archimago, they appear to be meeting that threshold, at least on average power.

 

[DimitryZ]

Isn't a simplified summary (for the first unfold) that the encoder is a band splitter, and the decoder joins the bands?
https://patents.google.com/patent/WO2017109495A1/en (Lossless bandsplitting and bandjoining using allpass filters)

Yeah!

 

[bennetng]

Just to comment about the DXD noise issue. DXD is nothing but a marketing name of 24-bit 352.8k PCM, therefore the noise is not inherent to the format itself. Download the raw DXD file here and see the noise floor decays in a similar way as shown on Merging's website.
http://www.2l.no/hires/DXD-DSD/index.html

https://merging.com/products/interfaces
No one prohibits interface manufacturers to add post-filtering at the ADC output to trim the noise, or recording engineers doing similar things in DAWs, or better, reduces to 44.1k and cuts useless stuff. In the case of piano solo in the raw DXD demo above, it makes perfect sense.

 

[DimitryZ]

Just to comment about the DXD noise issue. DXD is nothing but a marketing name of 24-bit 352.8k PCM, therefore the noise is not inherent to the format itself. Download the raw DXD file here and see the noise floor decays in a similar way as shown on Merging's website.
http://www.2l.no/hires/DXD-DSD/index.html

https://merging.com/products/interfaces
No one prohibits interface manufacturers to add post-filtering at the ADC output to trim the noise, or recording engineers doing similar things in DAWs, or better, reduces to 44.1k and cuts useless stuff. In the case of piano solo in the raw DXD demo above, it makes perfect sense.

Of course. That what MQA does.

However, if you are in the business of selling DXDs to the crazy audiophiles, they will not buy it from you if they hear it's been filtered.

 

[raistlin65]

Just ignore their advertising and irritation/desire to argue with it will dissipate over time.

I try to never see advertising and if forced to, to not pay attention.

That might work for you, but I help out with audio equipment recommendations almost everyday over on Reddit. And Tidal or MQA is often part of the conversation. Sometimes several times a week.

Now if you can persuade all of those people to ignore MQA's advertising, that sounds great. But I'm pretty sure you are not capable of that.

 

[ebslo]

So while MQA is not a completely lossless codec for all possible types of audio signals, it might indeed be lossless for most real world content.

MQA is not "lossless" under the standard definition for any high-resolution digital source material.

 

[bennetng]

Of course. That what MQA does.

However, if you are in the business of selling DXDs to the crazy audiophiles, they will not buy it from you if they hear it's been filtered.

MQA has nothing to do with this, don't attempt to hijack my comment. In the case of "crazy audiophiles", MQA can't fulfill their needs either. They would rather keep all the noise in the original DXD files, or filter them with their own methods, for example, with DSP utilties in software players.