MQA disappointing

I think you mean up till 48kHz no? Technically it is not because again it “borrows” bits from below the proviberal “noise floor” - those bits are lost forever. That said, it is here where “perceptually lossy” makes the most sense from a conceptual/theoretical point of view…

MQA core retrieve all the musical details at 88.2/96kHz. The renderer which ’ unfold’ further to 176.4/192kHz is actually up-sampling with final MQA filter to correct some timing issues to match the physical end of a DAC chip.

MQA authenticated at 192kHz is not the same to FLAC 192kHz when it comes to signal retrieval. The later may contain information above the 48kHz (192k sample).

2 Likes

Just a minor correction: between 44.1/48 & 88.2/96 MQA is lossy. Above 96 there is no signal at all - it’s just a play on words - it’s simply upsampling (which unless you have an exotic NOS and/or vintage DAC, was going to happen anyways).

As far as 16 bit is concerned, I believe MQA is likely meet the ‘lossless’ category even though it is no longer bit perfect. The reminder 8 bit to code above the audio range is lossy and a coarse approximation. The reproduction of ultrasounds is more on timing rather than details information.

The coding information embedded under the noise floor is used for unfolding; unfold 44.1/48kHz to 88.2/96kHz (MQA core). This process is based on approximation, it is not bit perfect either.

MQA is only good for sample equal or not greater than 88.2/96kHz. Anything above is a total stop band. Again the process is based on approximation, it is not bit perfect.

It is indeed a bit reduction from 24 down to 16 bit and according to him, 16 bit is more than enough to reproduce the required dynamic range in the audio range (CD). He also said we need the ultrasounds range to reproduce the precise timings. I think he had an diagram that showed how much it can encode the ultrasounds range… and it is around 48kHz (based on 96kHz sample).


After some noise shaping the result is 17 bit 96kHz (MQA core output). The diagram showed bit and sample splitting (encoder, 7A) and bit and sample stacking (decoder, 7B). There’s no evidence of psychoacoustic compression in the first 16 bit. The rest of bits are used as lossy approximation to re-construct the ultrasounds to 48kHz.

1 Like

The terms/definitions ‘lossy’ and ‘lossless’ are unambiguous.
‘Lossy’ means that information is irretrievably lost from the original data-stream/file, as part of a processing/compression process.
‘Partial lossy’ is just BS.
MP3 is lossy.
And so is MQA. MQA is ‘lossy’. No debate.

Thanks, these are answers that will help us understand things.

Say for example, if you pass a PCM to a DSP, either it is for upsampling from 48kHz to 192kHz or simply doing some room correction, this is considered an ‘enhancement’ rather than ‘lossy’ since the information is changed and no longer bit perfect from the original data as part of DSP process. Does this mean we are losing information? The answer is No!

MQA is lossless for the first ‘16 bit’, this where all details information is fully captured in the audio range, this is similar to the CD standard, if you argue that CD is indeed lossy? Now the next 8 bit is used to code information in the ultrasonic range up to 48kHz. There’s very little details, sonically above that range and mostly are timing information. 8 bit was chosen to give some information retrieval without increasing the overall bandwidth. MQA is just like another DSP and If you view that way, the course is not bad after. MP3 is based psychoacoustic principles; it throw away information based on our limit of hearings but in most case not everyone will hear the same way and this is problematic.

DSP’ing is transient.

To make your analogy maybe more relevant, if I took an EAC verified FLAC file, convolved it to sound good in my car, and then gave it to you, still a FLAC, to play on your home system, would you consider it “enhanced” or “lossy” in the definition that @Martin_Kelly used ?

Technically, no change of data.

If one assumes I am magnanimous and give you the convolution file, which is why I used that example rather than EQ.

If you taking about data integrity, bit perfect is necessary but it only applied where you need the exact information for data analysis. In audio, even it is not bit perfect, musical information is not necessary lost. It all depends on type of process you are taking about. For example, if you have 16/44.1 file and sample rate to 16/48kHz, you don’t lose information just because it is no longer bit perfect. That’s my argument.

MusicFidelity.
To me the discussion is closed thanks to you. In the sheet (see: red area) here the company is just honest but confusing for our discussion, they literally say ‘’ Lossy decompression + touchup to Lossless ‘’ Only in the Lossless explanation is still unclear whether the moved information is permanently gone?

The ‘lossy’ decompression that you indicated in red is for the lower 8 bit which carried information above the audio range. As far as 8 bit is concerned any touch up from ‘lossy’ to ‘lossless’ is based on the amount of information it exists. If more information exist and 8 bit is just not enough to code, it will revert to ‘lossy’ otherwise it is ‘lossless’ based on 8 bit code.

Lossless for 8 bit is easy to achieve, you only need 256 data points compared to 16 bit 65536 data points. This is very coarse approximation!

1 Like

You’re pretty much repeating BS’s argument, and that’s fine. It isn’t worth my time arguing it further. Your 16/44 => 16/48 example is specious and diversionary, inasmuch as I’ll assume you know as well as I do that it depends on how it’s done.

I would consider you an audiophile purist; demand no change of data, bit perfect and lossless way to enjoy your music and that’s perfectly fine. I ditched Tidal MQA in favour of Qobuz Hi-Res, not because it is lossy or partially lossless or lossless, but I don’t like the sound signature of MQA.

It’s more about language than it is about audiophilia TBH: lossy has a meaning, so does lossless. That put aside, I’m a huge believer in DSP.

There’s also a question of permanence: I strongly believe that using computational power to enhance sound quality is the future, especially at the speaker level, and that the possibilities of enhancement will evolve with time, in part thanks to better algos, and in part because we’ll be able to throw more processor power at the problem. This requires that the original files are as untouched as possible.

3 Likes

What a I hear when listening to MQA is lossy distortion. It is subtle but an experienced trained ear can hear it quite easily when comparing to the original lossless file.

DSP is great because a bad room can’t always be fixed. And a terrible speaker can sound a bit better with some DSP.

That said, there is nothing that replaces a high quality speaker that gives pristine low distortion flat uncoloured evenly dispersed response at all SPL levels.

That said, there is nothing that replaces a properly designed and acoustically treated room!

DSP is an excellent band-aid but it isn’t perfect.