MQA Discussion

[Kingsrule]

Meridian users..

I have a brand new 818v3 upgrade kit (not sure if it applicable to 808) available

Also 818v2 and several Rack Mount Twinstores available

PM if interested

 

[Fitzcaraldo215]

Certainly , DRC will not work in the way we do it nowdays
My DIRAC or Acourate or simple PEQ RC wont work at all as far as I can see

Depends on the specific products developed and marketed. MQA, by the way, will work straightforwardly with your existing equipment and PC software at 44k/24 bit. It just wil not offer all the time domain advantages of MQA, only half of them. The problem is retrieving hi rez encoded and embedded in that data.

If, a big IF, they offer the "phase 1" piece as computer software - a plug-in type of app, say - then that would allow you to decode - unzip - the MQA data to normal hi rez PCM, which can then undergo DSP EQ, bass management, etc. with your existing gear and EQ software in the PC. That unzipped data will have been MQA processed to eliminate the a-d time domain distortions on the recording side = half the claimed sonic advantages of MQA.

But, to gain full advantage of MQA, the other half of the puzzle which are time domain filter corrections for the playback side d-a, you need the "phase 2" MQA processing, which will require in an MQA-capable DAC. But, that is optional, and an ordinary DAC would still work, still delivering the recording side half of the corrections.

I know it is confusing. So, to review, MQA offers three basic things, all proprietary (for now):

1. A PCM data compression scheme, where hi rez is "transparently" encoded in a 44k/24 bit data stream (normally, but other resolutions are possible) which is compatible with most existing equipment, software, etc. for playback at 44/24 without any MQA processing. Retrieval of the hi rez data from this stream requires MQA Phase 1 processing. An MQA DAC will do that, but it is at least theoretically possible to perform this ahead of DSP EQ, etc. in a PC, much like FLAC, MP3 and many other compression codecs now work in PCs.

2. Correction for time domain filtering by a-d done on the recording side. This processing was done prior to distribution and it is built into the MQA data stream as received and is compatible with existing playback equipment.

3. Correction for time domain filtering by d-a done on the playback side. This requires MQA Phase 2 processing in an MQA DAC, but that is optional. All sampling rates are supported, regardless of whether or not an MQA DAC is used.

So, stay tuned. There is much more to learn as specific MQA products and media emerge. There is absolutely no panic at this point to reconfigure or upgrade our systems for MQA.

 

[jkeny]

Hi Amirim,

Though some might disagree with me, I would never called "perceptually lossless" lossless. Lossless means you don't lose anything -- and in this case you do. Now, one could argue that you can't tell, but it is still, on a bit-for-bit comparison, not truly lossless, which is what I was talking about. And if you encode perceptually lossless over and over again, you'll lose more and more. So I think using the term lossless is misleading. If something is grey, it's not black or white -- it's grey. It's either truly lossless or not.

For those wondering what "perceptually lossless" means, this video (which uses it in a video context) gives a good enough explanation.

Doug Schneider
SoundStage!

If the psychoacoustic model for "perceptually lossless" is accurate then it won't degenerate the perceived sound quality, no matter how many times it is encoded.

 

[jkeny]

Depends on the specific products developed and marketed. MQA, by the way, will work straightforwardly with your existing equipment and PC software at 44k/24 bit. It just wil not offer all the time domain advantages of MQA, only half of them. The problem is retrieving hi rez encoded and embedded in that data.

If, a big IF, they offer the "phase 1" piece as computer software - a plug-in type of app, say - then that would allow you to decode - unzip - the MQA data to normal hi rez PCM, which can then undergo DSP EQ, bass management, etc. with your existing gear and EQ software in the PC. That unzipped data will have been MQA processed to eliminate the a-d time domain distortions on the recording side = half the claimed sonic advantages of MQA.

But, to gain full advantage of MQA, the other half of the puzzle which are time domain filter corrections for the playback side d-a, you need the "phase 2" MQA processing, which will require in an MQA-capable DAC. But, that is optional, and an ordinary DAC would still work, still delivering the recording side half of the corrections.

I know it is confusing. So, to review, MQA offers three basic things, all proprietary (for now):

1. A PCM data compression scheme, where hi rez is "transparently" encoded in a 44k/24 bit data stream (normally, but other resolutions are possible) which is compatible with most existing equipment, software, etc. for playback at 44/24 without any MQA processing. Retrieval of the hi rez data from this stream requires MQA Phase 1 processing. An MQA DAC will do that, but it is at least theoretically possible to perform this ahead of DSP EQ, etc. in a PC, much like FLAC, MP3 and many other compression codecs now work in PCs.

2. Correction for time domain filtering by a-d done on the recording side. This processing was done prior to distribution and it is built into the MQA data stream as received and is compatible with existing playback equipment.

3. Correction for time domain filtering by d-a done on the playback side. This requires MQA Phase 2 processing in an MQA DAC, but that is optional. All sampling rates are supported, regardless of whether or not an MQA DAC is used.

So, stay tuned. There is much more to learn as specific MQA products and media emerge. There is absolutely no panic at this point to reconfigure or upgrade our systems for MQA.

One other aspect I think you missed:
- MQA is stated to deal with noise floor modulation (on both A/D & D/A sides?) but is not saying much about this, claiming some proprietary techniques but has stated that they may produce paper on this aspect at a later stage. I guess temporal-blurring is an easier sell at the moment to Joe Public - noise floor modulation requires too much explanation

 

[jkeny]

Very good article! Answering Doug's question, MQA is perceptually lossless. Not mathematically. If done perfectly, it should be equiv. to mathematically lossless.

I don't understand what you mean here, Amir - how can perceptually lossless be equiv to mathematical lossless - the "lossless" aspect of each are judged in totally different ways?

 

[Fitzcaraldo215]

Hi Amirim,

Though some might disagree with me, I would never called "perceptually lossless" lossless. Lossless means you don't lose anything -- and in this case you do. Now, one could argue that you can't tell, but it is still, on a bit-for-bit comparison, not truly lossless, which is what I was talking about. And if you encode perceptually lossless over and over again, you'll lose more and more. So I think using the term lossless is misleading. If something is grey, it's not black or white -- it's grey. It's either truly lossless or not.

For those wondering what "perceptually lossless" means, this video (which uses it in a video context) gives a good enough explanation.

Doug Schneider
SoundStage!

I disagree. I have no problem at all understanding the difference between "perceptually lossless" and "lossless". The meanings are abundantly clear, as is the fact that the two are not the same, though arguably they might achieve the same end result to listeners.

Yes, "lossless" is an absolute, unambiguous term. "Perceptually lossless" relies on research into psychoacoustic perception. You can quibble with that research. You can quibble offering your own listening findings whether MQA is truly lossless or not in your own perception. But, quibbling about the words themselves is pointless. It is even "perceptually pointless".

 

[das]

If the psychoacoustic model for "perceptually lossless" is accurate then it won't degenerate the perceived sound quality, no matter how many times it is encoded.

Hi,

I don't agree.

Many people can't hear the difference between a 320kbps MP3 and 16/44.1 file. Does that make MP3 "perceptually lossless." For some, yes.

Regardless. Perceptually lossless is, in fact, lossless to a degree, and each encoding will involve more and more degradation.

Take image turning. I'm not sure how the current operating systems work and if it's still valid, but under Windows 7 you could turn an image and it would look the same -- even if it wasn't quite the same. The degradation wasn't noticeable. A few more times and you were still fine. After a certain number of turns, the subtle degradations with each turn started to show and you had a grossly distorted image. It was probably "perceptually lossless" one turn to the next, but not one turn to the 100th or 1000th turn.

I suspect the same will happen with any lossy system. On the other hand, convert WAV to FLAC and FLAC to WAV all day long and then some, and you'll always wind up the same.

Doug Schneider

 

[amirm]

Hi Amirim,

Though some might disagree with me, I would never called "perceptually lossless" lossless. Lossless means you don't lose anything -- and in this case you do. Now, one could argue that you can't tell, but it is still, on a bit-for-bit comparison, not truly lossless, which is what I was talking about. And if you encode perceptually lossless over and over again, you'll lose more and more. So I think using the term lossless is misleading. If something is grey, it's not black or white -- it's grey. It's either truly lossless or not.

Hi Doug. We are actually in agreement. You are describing what is formally called mathematically lossless. The explanation I added is another type which is perceptually lossless. It means it is lossless to ear. Not lossless to the bits.

If you use a lossy codec in variable bit rate (VBR) mode and tell it to use max quality, then you have a perceptually lossless codec. The codec will use its psychoacoustic model of human hearing and only discard data that keeps the overall fidelity below threshold of audibility. If done right, it is extremely hard to tell the input from output. Trained listeners may still be able to spot fidelity problems because their hearing system operates above that of general public where the psychoacoustic models are built. In that sense, they still miss the bar with audiophiles.

In the case of MQA as I think you alluded to in your article, the job is a lot easier for a perceptual lossless codec. For one thing, we can't hear anything above 20 Khz so whatever the codec does, not matter how broken in that region, will likely still not matter! MQA preserves the data above noise there for the sake of completeness. Accuracy is paramount in the audible range and if there MQA maintains lossless fidelity, it is already there. So in that sense, MQA is a hybrid mathematically lossless+perceptual lossy coder.

Thanks again for a good article and skeptic point of view .

 

[amirm]

Many people can't hear the difference between a 320kbps MP3 and 16/44.1 file. Does that make MP3 "perceptually lossless." For some, yes.

This is fixed rate encoding. The codec is told to maintain 320 kbps where it can achieve transparency or not. In other words, priority is placed on bit rate, not fidelity.

As I explained, the proper example is to use variable bit rate where quality is high priority, and the codec is free to use whatever bit rate it wants, all the way up to 1.4 mbit/sec of CD.

Also, MP3 was never designed for transparency. A better example is to use AAC in VBR mode with max quality. Vast majority of audiophiles will fail this test. We did that with WMA Pro at Microsoft and indeed that was the outcome.

 

[jkeny]

Hi,

I don't agree.

Many people can't hear the difference between a 320kbps MP3 and 16/44.1 file. Does that make MP3 "perceptually lossless." For some, yes.

But you are using the phrase "perceptually lossless" incorrectly - if some people can hear a difference between 320kps MP£ & 16/44 then it isn't "perceptually lossless" - it's close but no cigar.

Regardless. Perceptually lossless is, in fact, lossless to a degree, and each encoding will involve more and more degradation.

This is incorrect as I already said - you are using the wrong definition.

 

[das]

Hi Doug. We are actually in agreement. You are describing what is formally called mathematically lossless. The explanation I added is another type which is perceptually lossless. It means it is lossless to ear. Not lossless to the bits.

Hi,

Yes, I do realize we're in agreement on this. I just wanted to put out that I find "perceptually lossless" to be an oxymoron -- at least insofar as audiophiles think about lossless. Lossless in the audiophile world has been synonymous with what you're referring to as mathematically lossless.

Thanks,
Doug

 

[das]

This is incorrect as I already said - you are using the wrong definition.

How exactly is this wrong about degradation increasing with subsequent encodes?

Doug

 

[jkeny]

How exactly is this wrong about degradation increasing with subsequent encodes?

Doug

I don't want to get into pedantic definitions but if a process is actually 100% "perceptually lossless" (which is what I understand it to mean) then it doesn't matter how many times the process is run iteratively on the same audio signal, it will always remain perceptually the same sound. But agreed that there might be degradation - but that degradation will be of no consequence, perceptually

 

[das]

I don't want to get into pedantic definitions but if a process is actually 100% "perceptually lossless" (which is what I understand it to mean) then it doesn't matter how many times the process is run iteratively on the same audio signal, it will always remain perceptually the same sound.

Hi,

Perhaps I wasn't clear enough. I'm talking about generation after generation -- the original file to a new file, then than new file into another file, and so on... The losses will compound.

Doug Schneider