MQA Rendering-- what is it?

[alfamatt]

Trying to get my head around MQA.

ive heard it in a few set ups with blue sound equipment, i.e. MQA enabled DAC.

At home, I'm trying Tidal-MQA Masters with Audirvana 3 but a with non MQA DAC. It's a Jolida glass DAC.

Audirvana 3 "renders" MQA. What does this mean in plain old English please? Reducing " time domain blurring" ? 

One thing I can say is that Tidal streaming at hifi sounds better through Audirvana than without it. But my old Mac Mini with 4gb RAM is struggling, telling me I've run out of memory, so if I'm going to take the plunge with the Audirvana/Tidal/MQA, it looks like I'm going to have to get more ram!

[miguelito]

Ok I'll bite...

 

Disclosure: I've read a bunch of stuff and this is how I've come understand this... Also please understand that I am not taking any sides here. I have listened to MQA and I like what I hear but I am also a skeptic when something has so much PR behind it.

 

MQA decoding is divided into two steps:

1- Unfolding to a 2x sampling rate:

Example: a 48KHz/24bit MQA file is decoded into an 96KHz/24bit file. The information for this decoding is stored in the 8 least significant bits in the 24bit samples. My interpretation is that the 16 most significant bits in the original 48/24 data are true to the source (ie lossless) whereas when you unfold you have a mix whereby the frequency range from 0-24KHz is true to the original (with a resolution/bit depth of 16bits give or take) and the remaining has lower bit depth (the argument being that the full bit depth is not needed given the ear's resolution). Not crazy, but certainly not mathematically lossless.

 

2- Rendering:

This step involves upsampling but with a twist: The decoded PCM stream includes information on the most appropriate upsampling parameters in some unified description which then needs to be translated into a specific DAC chip settings to be set. Think of the iZotope parameters in Audirvana as parameters in the DAC chip: The rendering stage simply sets the parameters on the DAC chip based on some more generic protocol in the MQA stream (which up until now has no idea what the DAC is). To give you an example: The MQA stream might say something like "transients=high" which the renderer interprets and then sets the DAC to a minimum phase filter rather than a linear phase one. The renderer needs to set these parameters in the DAC so it's clearly DAC specific.

 

Reiterating the disclosure: this is my interpretation/extrapolation of all of the stuff I have read.

[miguelito]

To clarify one thing: TIDAL and Audirvana 3 are only doing step 1 (unfolding), not step 2. It will skip step 1 if it detects/you tell it to do so.

 

If my interpretation of step 2 is correct, it might be possible in the future to translate the MQA metadata indicating the upsamplimg profile/signature into specific iZotope parameters automatically. Of course I don't expect MQA (the company) to reveal the upsampling signature language any time soon. But a lot of smart people can figure this out...

[revand]

Based on a previous article here I tried to make a visual explanation of MQA unfolding and rendering here.

[miguelito]

17 hours ago, revand said:

Based on a previous article here I tried to make a visual explanation of MQA unfolding and rendering here.

I have to tell you - I find your article confusing.

[revand]

22 hours ago, miguelito said:

I have to tell you - I find your article confusing.

 

I tried to be simple and straight forward

If you have any question just ask me...

[audiventory]

Here patent where described full coding/decoding system https://patentscope.wipo.int/search/en/detail.jsf?docId=WO2014108677&recNum=1&maxRec=&office=&prevFilter=&sortOption=&queryString=&tab=PCT+Biblio

 

There is term "rendering".

 

Real implementation may have difference with the patent.

[mansr]

On 4/15/2017 at 1:32 PM, miguelito said:

MQA decoding is divided into two steps:

1- Unfolding to a 2x sampling rate:

Example: a 48KHz/24bit MQA file is decoded into an 96KHz/24bit file. The information for this decoding is stored in the 8 least significant bits in the 24bit samples. My interpretation is that the 16 most significant bits in the original 48/24 data are true to the source (ie lossless) whereas when you unfold you have a mix whereby the frequency range from 0-24KHz is true to the original (with a resolution/bit depth of 16bits give or take) and the remaining has lower bit depth (the argument being that the full bit depth is not needed given the ear's resolution). Not crazy, but certainly not mathematically lossless.

The number of bits used for MQA data is actually not fixed. In practice, all files I've checked use the 8 least significant bits for the compressed high-frequency content. The 9th bit carries various metadata and cryptographic signatures. The high 15 bits contain the low frequencies, though not entirely unmolested. The low-pass filter used in the encoding has quite a bit of aliasing, and there is also some added pseudorandom noise which is removed by the decoder.

On 4/15/2017 at 1:32 PM, miguelito said:

2- Rendering:

This step involves upsampling but with a twist: The decoded PCM stream includes information on the most appropriate upsampling parameters in some unified description which then needs to be translated into a specific DAC chip settings to be set. Think of the iZotope parameters in Audirvana as parameters in the DAC chip: The rendering stage simply sets the parameters on the DAC chip based on some more generic protocol in the MQA stream (which up until now has no idea what the DAC is). To give you an example: The MQA stream might say something like "transients=high" which the renderer interprets and then sets the DAC to a minimum phase filter rather than a linear phase one. The renderer needs to set these parameters in the DAC so it's clearly DAC specific.

It's simpler than that. The metadata specifies only which filter number to use. It's a 5-bit field, allowing 32 different values, but only a handful are actually defined. Presumably they left some space for future additions.

 

All MQA filters are minimum phase.

[miguelito]

5 hours ago, mansr said:

The number of bits used for MQA data is actually not fixed. In practice, all files I've checked use the 8 least significant bits for the compressed high-frequency content. The 9th bit carries various metadata and cryptographic signatures. The high 15 bits contain the low frequencies, though not entirely unmolested. The low-pass filter used in the encoding has quite a bit of aliasing, and there is also some added pseudorandom noise which is removed by the decoder.

It's simpler than that. The metadata specifies only which filter number to use. It's a 5-bit field, allowing 32 different values, but only a handful are actually defined. Presumably they left some space for future additions.

 

All MQA filters are minimum phase.

See? Told you smart folk would figure it out. Each one of the 5 bits prob turns on/off a property. 

 

"Unmolested" is brill. Consider that stolen!

[mansr]

41 minutes ago, miguelito said:

See? Told you smart folk would figure it out. Each one of the 5 bits prob turns on/off a property.

No, it's just a number, at least as far as the decoder is concerned. If there's any system to the allocation of these numbers, I haven't found it.

[plissken]

Why do they keep calling it 'unfolding' when it's decompressing?

[mansr]

3 minutes ago, plissken said:

Why do they keep calling it 'unfolding' when it's decompressing?

Because marketing.

[miguelito]

9 minutes ago, plissken said:

Why do they keep calling it 'unfolding' when it's decompressing?

That's what MQA calls it, and it's not decompressing in the standard sense: approx the 16 most significant bits are almost identical to the original. This allows playback as-is without any further processing. The remaining 8 least significant bits are "unfolded" by some algorithm into the higher frequency information. 

 

I supposed you could call this a form of lossy

compression but it is very different from a zip-like algo that is used in say FLAC. 

[mansr]

14 minutes ago, miguelito said:

That's what MQA calls it, and it's not decompressing in the standard sense: approx the 16 most significant bits are almost identical to the original. This allows playback as-is without any further processing. The remaining 8 least significant bits are "unfolded" by some algorithm into the higher frequency information. 

 

I supposed you could call this a form of lossy compression but it is very different from a zip-like algo that is used in say FLAC. 

It's not very different from traditional lossy compression methods such as mp3 or Dolby. Most of them work by splitting the audio signal into a number of frequency bands, then processing each band in some way or other. MQA differs only in that the lowest band covers roughly 0-20 kHz (the audible range) and is mostly untouched by the compression.

[miguelito]

14 hours ago, mansr said:

It's not very different from traditional lossy compression methods such as mp3 or Dolby. Most of them work by splitting the audio signal into a number of frequency bands, then processing each band in some way or other. MQA differs only in that the lowest band covers roughly 0-20 kHz (the audible range) and is mostly untouched by the compression.

I think you meant "unmolested"