advertisement


MQA pt II

Zombie said:
Oh no, maybe you'll like it. Dangerous...
How would you handle your expectation bias? ;)
Click to expand...
On the "crazy fuse" thread, I described an experience that went against expectation bias, or inverted it. I spent some effort at building stuff with great expectations of sonic bliss and economic efficiency. Unfortunately, it sounded like crap. The threaders offered several potential psychological explanations.

JimA's experience may be similarly tricky.
 
Last edited:
Rosewind said:
Bad publicity is better than no publicity. Right.

Tidal still has MQA content, and both Roon and Audirvana support MQA for some reason. I was considering Audirvana because of its Qobuz integration but I really don't feel comfortable supporting a music player that is tied in any way to MQA, so I upgraded from JRiver 23 to 27/28 and now I have no crashes when going from Qobuz to JRiver and back - as long as I close Qobuz before starting kernel streaming on JRiver.

I would not mind having an MQA filter as an option on a DAC. Then I can decide if music files taken through the "secret origami sauce process" is something I like or not. I would also not have to deal with all the other consequences of having music files only available as MQA. As you all know, MQA-files can only be properly MQA-unfolded using proprietary software and/or hardware. Even then I don't get the full information of an untouched file but instead an altered music profile because the MQA process involves the addition of "anharmonic elements and deterministic patterns" that will translate as noise in DACs without (and with?) the MQA-decoder in them. I'm looking forward to Jimaudiomisc's results.
Click to expand...
By considering MQA a "filter" you misunderstand the entire effort.

And by quoting JimA's "anharmonic distortion" meme, you prove you don't understand him at all. He is only looking at MQA into non-MQA hardware.

And, JimA, here is someone who is completely misunderstanding your research, just as I feared.

Perhaps you will see that constant explanations/disclaimers are needed when you post, so as not to mislead the poets among us.

If/when you get an MQA DAC for testing and listening (5 years after system's commercial introduction), you can offer us more concrete observations and measurements.

I look forward to that time, though I hope it takes less than another 5 years. Unless mansr agrees to setup a listening/measurement MQA test in the meantime.
 
The quirky thing about an MQA music file is that it may sound better than LPCM to some people despite being a lossy format.

What exactly MQA does to the lossless source file to create that sense of overall goodness to some people is what we are trying to find out.

Also we are trying to establish a level playing field for an open evalustion of MQA vs LPCM. This means that we need to be sure that we evaluate MQA and LPCM on the basis of the same unmodified source files.
 
DimitryZ said:
He is only looking at MQA into non-MQA hardware.
Click to expand...
That is a good thing, me and I am sure there are others who actually more interested to see what damage is done when you play MQA file on standart equipment rather than what is fully decoded MQA.
 
DimitryZ said:
Jim, I understand that you are very proud of your passion for the scientific method and your new discoverings.

But you do understand that there is literally nothing new or novel in your writings (at least the ones you linked to)?

Correction:
On reflection, above is negative and unnecessary snarky. My point, expressed more neutrally, is that there are a multitude of data compaction techniques (FLAC, TIFF, MLP, etc.) and more data compression techniques (JPEG, AAC, MPEG, VC, etc) that have been developed over decades for sound, image and video applications. What you described in your links are foundational concepts that have long been incorporated into various free and commercial codecs.
Click to expand...

Agreed. :) That's one reason it puzzles me that you keep seeming to need reminding about my approach being well established. However pleased that you do understand.
 
DimitryZ said:
By considering MQA a "filter" you misunderstand the entire effort.

And by quoting JimA's "anharmonic distortion" meme, you prove you don't understand him at all. He is only looking at MQA into non-MQA hardware.
Click to expand...

Not quite correct, despite your earlier assurance that you understand these issues.

You can, indeed, consider the 'folding and unfolding' using the MQA 'lazy filters' as a form of filter. But the problem then is that some forms of alteration can be fully reversed whilst others can't. As I've pointed out.

Your second assertion depends on the definition of 'is'. Yes AT PRESENT I am looking only at the changes in terms of LPCM. But I do intend - as I have said more than once and keep referring back to - to then examine what emerges from real-world MQA 'decoding'. Indeed, to some extent I am already doing so given GOs examples of 'unfolding', albeit that isn't as satisfactory as being able to do that for myself.

So fear not, your patience should be rewarded. :)
 
One curio I've encountered is illustrated here for people to ponder.

http://jcgl.orpheusweb.co.uk/temp/Dispersion.png

This compares the four versions of the GoldenOne files sourced and sent to Tidal as a 44k/24 version. The section shown has an impulse function in the source GO sent. i.e. One non-zero sample. The bottom graph plots that against the 'MQA' and 'Tidal Master' versions then issued by Tidal. The surprise is that *both* exhibit having been run though quite a dispersive filter that spreads the impulse over more than 2ms. The upper graph shows the 'unfolded' result when GO played back the MQA version. (All IIUC what he has said about the files.) This also shows a similar dispersion.

This raises a puzzle:

It has been applied to *both* the outputs from Tidal - MQA and non-MQA. Hence anyone comparing them to decide if they liked MQA or not would find both of them affected by the dispersion. Which would effectively 'blurr' them both if we use MQA terminology. if this is general, then other Tidal output may also be affected. If this is an error somewhere, I'm not sure where. So someone else might like to check this by inspection. Note that the source file has nothing whatsoever at or above 22.05kHz. So MQA should do nothing - although the MQA 'unfolding' to 88k2 output adds an aliased image of that to the region which wasn't in the source. Hence anyone playing this may be better off to ensure they are NOT 'decoding' it.

Not at all sure why this seems to have happened, or if this dispersion is a norm for tidal. Odd however you look at it.

Note the lower plots all 'show the dots' so you can see the sample values. The upper plot draws a line to give some idea of the waveform, but note this is a 'Legato Link' waveform not exactly what a standard sinc conversion would produce.

Interested to see what people make of it. Thought people might find it interesting as I delve further.
 
Ah! Thanks, that can explain it. And also why the flac files I started from have different sizes. That was puzzling given the results. I wondered if it was different tagging in the headers. (I don't bother with that in general.)

So the implication becomes that it is MQA that is 'blurring' the impulse function and then not actually 'deblurring' it when 'unfolded'? If so, not what people might expect to be happening. Maybe a part of the 'shapes' is aimed at correcting this in some other way like a time-reverse of the applied function. But if so it means that if 'blur' matters then you get 'added blur' if you want to play the files without MQA. Not good for anyone wanting to compare MQA with non-MQA by any means, including listening. In effect you'd get a 'degraded' version as 'non MQA' to compare with MQA.

Nor is reducing them to 12 bits - particularly if they then use bits for genuine 'HF' thus raising the (apparently) noise floor. However when I do a sample 'diff' the result is all zeros. Which implies 12 bits for both versions, otherwise you'd get the residual in the diff output. I get all zeros.

If the above is correct the question becomes: is Tidal doing this or do MQA require them to?
 
Actually that remains a puzzle given that I get all zeros from a diff. It implies the files are absolutely identical. Maybe I was unlucky when I did the diffs on randomly chosen sections. I'll also do some stats on the files to see the probability distributions of sample values. That would show up excess quantisation. This starts to seem like 'down the rabbit hole' rather than 'wonderland'...
 
Sorry, what I wrote wasn't quite complete. The files made from the 44.1 kHz input contain identical data:
Code: 
$ metaflac --show-md5sum */441khz.flac
MQA Encoded/441khz.flac:8ca3e615ee025ef7b97d049f337da8fa
Tidal HiFi version/441khz.flac:8ca3e615ee025ef7b97d049f337da8fa

The FLAC metadata differs, however:
Code: 
$ soxi */441khz.flac

Input File     : 'MQA Encoded/441khz.flac'
Channels       : 2
Sample Rate    : 44100
Precision      : 16-bit
Duration       : 00:08:00.47 = 21188863 samples = 36035.5 CDDA sectors
File Size      : 34.0M
Bit Rate       : 566k
Sample Encoding: 16-bit FLAC
Comments       :
ENCODER=MQAEncode v1.1, 3091 (afa6eeb9), F8EC1703-7616-45E5-B81E-D60821434062, Apr 09 2021 01:15:11
MQAENCODER=MQAEncode v1.1, 3091 (afa6eeb9), F8EC1703-7616-45E5-B81E-D60821434062, Apr 09 2021 01:15:11
ORIGINALSAMPLERATE=44100
ALBUM=The Callout
TRACKNUMBER=1
TRACKTOTAL=1
TITLE=The Callout
DISCNUMBER=1
DISCTOTAL=1
COPYRIGHT=GoldenSound Records
ALBUMARTIST=GoldenSound
ARTIST=GoldenSound
DATE=2021


Input File     : 'Tidal HiFi version/441khz.flac'
Channels       : 2
Sample Rate    : 44100
Precision      : 16-bit
Duration       : 00:08:00.47 = 21188863 samples = 36035.5 CDDA sectors
File Size      : 34.0M
Bit Rate       : 566k
Sample Encoding: 16-bit FLAC
Comments       :
ALBUM=The Callout
TRACKNUMBER=1
TRACKTOTAL=1
TITLE=The Callout
DISCNUMBER=1
DISCTOTAL=1
COPYRIGHT=GoldenSound Records
ALBUMARTIST=GoldenSound
ARTIST=GoldenSound
DATE=2021

Notice that the "hi-fi" version is missing the ENCODER and MQAENCODER fields. This means that any software that simply looks for these comment fields will misidentify the "hi-fi" file as not being MQA coded. I can't see how they could have done this with anything other than deceptive intentions.

The files sourced from the 88.2 kHz originals differ in bit depth, one being coded as 24-bit and the other as 16-bit:
Code: 
$ soxi */882khz.flac

Input File     : 'MQA Encoded/882khz.flac'
Channels       : 2
Sample Rate    : 44100
Precision      : 24-bit
Duration       : 00:05:00.89 = 13269032 samples = 22566.4 CDDA sectors
File Size      : 63.4M
Bit Rate       : 1.69M
Sample Encoding: 24-bit FLAC
Comments       :
ENCODER=MQAEncode v1.1, 3091 (afa6eeb9), F8EC1703-7616-45E5-B81E-D60821434062, Apr 09 2021 01:32:41
MQAENCODER=MQAEncode v1.1, 3091 (afa6eeb9), F8EC1703-7616-45E5-B81E-D60821434062, Apr 09 2021 01:32:41
ORIGINALSAMPLERATE=88200
ALBUM=The Callout (Deluxe)
TRACKNUMBER=1
TRACKTOTAL=1
TITLE=The Callout (Deluxe)
DISCNUMBER=1
DISCTOTAL=1
COPYRIGHT=GoldenSound Records
ALBUMARTIST=GoldenSound
ARTIST=GoldenSound
DATE=2021


Input File     : 'Tidal HiFi version/882khz.flac'
Channels       : 2
Sample Rate    : 44100
Precision      : 16-bit
Duration       : 00:05:00.89 = 13269032 samples = 22566.4 CDDA sectors
File Size      : 23.7M
Bit Rate       : 631k
Sample Encoding: 16-bit FLAC
Comments       :
ALBUM=The Callout (Deluxe)
TRACKNUMBER=1
TRACKTOTAL=1
TITLE=The Callout (Deluxe)
DISCNUMBER=1
DISCTOTAL=1
COPYRIGHT=GoldenSound Records
ALBUMARTIST=GoldenSound
ARTIST=GoldenSound
DATE=2021[CODE]

Looking at the samples of the "hi-fi" file, we see that only 12 bits are actually used:
[CODE]
 sox Tidal\ HiFi\ version/882khz.flac -n stats
             Overall     Left      Right
DC offset  -0.000043 -0.000019 -0.000043
Min level  -1.000000 -1.000000 -1.000000
Max level   0.999512  0.999512  0.999512
Pk lev dB       0.00      0.00      0.00
RMS lev dB    -15.69    -15.26    -16.16
RMS Pk dB      -4.52     -4.52     -5.81
RMS Tr dB     -20.23    -20.17    -20.23
Crest factor       -      5.80      6.42
Flat factor     0.00      0.00      0.00
Pk count          43        77         9
Bit-depth      12/12     12/12     12/12
Num samples    13.3M
Length s     300.885
Scale max   1.000000
Window s       0.050

These 12 bits are equal to the top 12 bits of the "master" version.
 
Here is another aspect of the above.

http://jcgl.orpheusweb.co.uk/temp/MQAInterval.png

Given the comments made by Mansr I ran my Flac checking stats prog to get that. The values plotted need a bit of explaining. The 'interval' values here stem from:

a) Count up how often every possible sample value occurs in the scanned span of a file. (Here set to a max of 5 mins.)

b) Then FFT that set of values to get a 'spectrum of the intervals'. If, for example, there were more occurrances of values of 8, 16, 24, etc, then you'd see a peak at 8 in the 'spectrum'.

The curio here for the MQA file is the shape in the region below and interval of 10 or so - with a distinct 'dip' at about 6.
Plus the shape overall is very different to a randomly chosen comparison (Scholl singing some Handel.) The reason for this is unknown, but it is odd. So I need to spend time checking the other versions, etc. Might be from source.
 
Jim Audiomisc said:
Here is another aspect of the above.

http://jcgl.orpheusweb.co.uk/temp/MQAInterval.png

Given the comments made by Mansr I ran my Flac checking stats prog to get that. The values plotted need a bit of explaining. The 'interval' values here stem from:

a) Count up how often every possible sample value occurs in the scanned span of a file. (Here set to a max of 5 mins.)

b) Then FFT that set of values to get a 'spectrum of the intervals'. If, for example, there were more occurrances of values of 8, 16, 24, etc, then you'd see a peak at 8 in the 'spectrum'.

The curio here for the MQA file is the shape in the region below and interval of 10 or so - with a distinct 'dip' at about 6.
Plus the shape overall is very different to a randomly chosen comparison (Scholl singing some Handel.) The reason for this is unknown, but it is odd. So I need to spend time checking the other versions, etc. Might be from source.
Click to expand...
There are numerous other comparisons you can run.

The field of numerology is vast.

:)
 
Jim Audiomisc said:
Not quite correct, despite your earlier assurance that you understand these issues.

You can, indeed, consider the 'folding and unfolding' using the MQA 'lazy filters' as a form of filter. But the problem then is that some forms of alteration can be fully reversed whilst others can't. As I've pointed out.

Your second assertion depends on the definition of 'is'. Yes AT PRESENT I am looking only at the changes in terms of LPCM. But I do intend - as I have said more than once and keep referring back to - to then examine what emerges from real-world MQA 'decoding'. Indeed, to some extent I am already doing so given GOs examples of 'unfolding', albeit that isn't as satisfactory as being able to do that for myself.

So fear not, your patience should be rewarded. :)
Click to expand...
In a poet's world, a "filter" has a more common understanding than can be applicable in engineering.

And a good engineer always understands their audience.
 
You must log in or register to reply here.


advertisement


Back
Top