Great! Now explain how MQA uses compressed sensing and how audio signal satisfy the sparcity and incoherence criteria required for compressed sensing to work.
Tom
If you look at the papers the MQA cited in reference to what has published about how MQA works, IIRC, there is a specific reference to compressed sensing. And there are also references to a lot other modern theory.
The way I look at it is this: music produced by acoustic instruments does not contain high levels of high frequency harmonics. That means there is statistically underutilized channel capacity as viewed in the frequency domain. If you understand that if there is any useful domain someone can think up wherein musical signal information can be represented by sparse equations, there is an opportunity to find optimal reconstruction solutions to undersampled data. If undersampled data can be successfully recovered by such means, then then there may be an opportunity to utilize otherwise underutilized capacity in a channel. That is exactly one of the places MQA hides additional information: In underutilized areas of the frequency versus intensity level space. IOW, there is room to encode more in a 16/44 channel (if it is cleverly encoded) in mostly unused frequency/intensity_level space at higher audio frequencies where acoustic instruments and voices tend to produce only low level harmonics.
Also to be very clear, the above is not to approve or defend what MQA does, its only in relation to some attempt to understand how it could approach some of the problems involved in doing what it claims to do. My personal opinion is that I don't like MQA and I am happy to see it gone.
The way I look at it is this: music produced by acoustic instruments does not contain high levels of high frequency harmonics. That means there is statistically underutilized channel capacity as viewed in the frequency domain. If you understand that if there is any useful domain someone can think up wherein musical signal information can be represented by sparse equations, there is an opportunity to find optimal reconstruction solutions to undersampled data. If undersampled data can be successfully recovered by such means, then then there may be an opportunity to utilize otherwise underutilized capacity in a channel. That is exactly one of the places MQA hides additional information: In underutilized areas of the frequency versus intensity level space. IOW, there is room to encode more in a 16/44 channel (if it is cleverly encoded) in mostly unused frequency/intensity_level space at higher audio frequencies where acoustic instruments and voices tend to produce only low level harmonics.
Also to be very clear, the above is not to approve or defend what MQA does, its only in relation to some attempt to understand how it could approach some of the problems involved in doing what it claims to do. My personal opinion is that I don't like MQA and I am happy to see it gone.
Last edited:
Appreciate the explanation, though it didn't really address my question.
If MQA's claims that they leave the 16 MSBs untouched in their process, it should be easy to confirm by running a track through the MQA encoder and comparing the output file with the input file. MQA hasn't shown this and third-party attempts to do so have either been prevented by MQA or have shown that the MQA process does alter the 16-bit data, thus, making MQA a lossy form of encoding.
I just don't see a value in yet another lossy audio file format.
Tom
If MQA's claims that they leave the 16 MSBs untouched in their process, it should be easy to confirm by running a track through the MQA encoder and comparing the output file with the input file. MQA hasn't shown this and third-party attempts to do so have either been prevented by MQA or have shown that the MQA process does alter the 16-bit data, thus, making MQA a lossy form of encoding.
I just don't see a value in yet another lossy audio file format.
Tom
HiRes FLAC has lossy payload to start with especially if downconverted from original studio master. At least in MQA, the original master is used but some losses in the process that may not be that audible in the end…
Acko, FLAC is lossless, you can run a file compare of the flac file after complementary compression and expansion and you will find zero difference between it and the original PCM file. (FLAC can support 32 bit 768k sample rate material) Reduced resolution in FLAC files has nothing to do with the FLAC codec, and unlike MQA supports PCM without perceptual coding or other tricks.)
You can use Audacity or bit compare to make the comparison. For encoding/decoding flac files for testing you can use FLAC front end - not sure whether it supports modern OS, but there may be alternatives. ( I used this early on to confirm the flac encoding was lossless - way back in the late WinXP era.)
https://flacfrontend.sourceforge.net/
https://forum.audacityteam.org/t/comparing-two-supposedly-identical-tracks/36424
MQA was developed by Meridian and released in 2014, I think it was late to the game as it addressed bandwidth and file size issues that were no longer a big deal even then. It's allure was 44/16 file sizes offering higher resolution. There are MQA CDs and players that can play them, again rather late to the game.
Only abou 5% of the music I listened to on Tidal was MQA, in certain categories you will find some 2496 flac encoded files, but a great deal of material is still redbook. Tidal also substantially reduced their subscription rates and as a result I expect a lot of the MQA files to be replaed with redbook quality files and not high res. I have noticed that the quality of a lot of redbook material sounds better today than I remember in the pas, and I listen to it on a system that is pretty revealing.
You can use Audacity or bit compare to make the comparison. For encoding/decoding flac files for testing you can use FLAC front end - not sure whether it supports modern OS, but there may be alternatives. ( I used this early on to confirm the flac encoding was lossless - way back in the late WinXP era.)
https://flacfrontend.sourceforge.net/
https://forum.audacityteam.org/t/comparing-two-supposedly-identical-tracks/36424
MQA was developed by Meridian and released in 2014, I think it was late to the game as it addressed bandwidth and file size issues that were no longer a big deal even then. It's allure was 44/16 file sizes offering higher resolution. There are MQA CDs and players that can play them, again rather late to the game.
Only abou 5% of the music I listened to on Tidal was MQA, in certain categories you will find some 2496 flac encoded files, but a great deal of material is still redbook. Tidal also substantially reduced their subscription rates and as a result I expect a lot of the MQA files to be replaed with redbook quality files and not high res. I have noticed that the quality of a lot of redbook material sounds better today than I remember in the pas, and I listen to it on a system that is pretty revealing.
Last edited:
Thanks! I was referring to the ‘payload’ not FLAC itself which is indeed lossless. Tidal’s new ‘Max’ FLAC subscription is not very convincing to continue the same way as I used to with MQA…. because not sure what is being transported in the FLAC container. I have to find out more
Generally my experience with Tidal and sound quality over the past 3 years, whether MQA or other formats, has been good. I have run across recordings in all formats I did not think sounded good on Tidal and elsewhere.
The claim that the short impulse response stuff is the interesting part is not a claim but an opinion, my opinion. Apparently the new owners of MQA hold the same opinion, if I understand this link that acko dug up correctly https://www.headphonesty.com/2024/06/mqa-returns-airia-foqus-qrono/
Peter Craven provided only anecdotical evidence for the idea that short impulse responses are useful for humans in his article about apodizing filters. In the article "A hierarchical approach to archiving and distribution" by Bob Stuart and Peter Craven, AES convention paper 9178,
https://aes2.org/publications/elibrary-page/?id=17501 , they provide references to all sorts of interesting articles, about alternative perfect reconstruction sampling theorems for non-bandlimited signals with a finite rate of innovation for example, but not much about perception of short impulse responses.
In fact, in "A hierarchical approach" they claim that humans might be able to hear the envelope of ultrasonic signals and then refer to two references, [86] and [90]. When you look up the abstracts, one is about barn owls and the other about cats.
The proof is there... see below... not that I needed it when I always preferred AKM silicon to early ESS.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8261637/
"It is generally believed that ultrasound cannot be heard. However, ultrasound is audible when it is presented through bone conduction....."
"In previous studies on ultrasonic perception at the central level, neural activation was observed with magnetoencephalography and positron emission tomography in the auditory cortex but not in the somatosensory cortex [9,10]. These findings objectively established that ultrasonic perception is an auditory sensation, not a somatosensory sensation."
... not only it is audible - but the mechanism is implemented, through clinical research, with real-life implementations/benefits.
... a must-read article.
https://drive.google.com/file/d/1E7t0JxqlVmTuoLk-1cYAvopRkbrh48jC/view?usp=sharing
Last edited:
I dont think so really - not the "compressor" for the flac coder - it is was transparent as .zip - if you believe that zip gives back your files, so will flac.
Then we could discuss what is put into these coders but I think thats not really the point - right?
//
Haha - there is no modern sampling theory, Shannon and Nyqvist is what it is - no news here - sorry mate.
//
There are actually more perfect reconstruction sampling theorems, the one for impulse sampled bandlimited signals is just the most well-known one. You also have such theorems for pulse width modulation instead of impulse sampling, and for impulse sampling non-bandlimited signals with a finite rate of innovation.
Thanks for the education!
How about the new UWB (ulttra wideband) solution that everybody seems to write about?
Jan
Com'on, the "layer" below 16 bits in MQA was coded as "mp3" i.e. lossy coding. Just the idea to stitch differently coded layers together is insane.... really unprofessional. I cant see that any one technical insightful would find any beauty in this wreck of an audio standard.
The whole thing is a scam as I see it.
//
Yes there are other but I see the "old ones" as flawless really. Realisation is an other matter.. 🙂
//
- Status
- Not open for further replies.