advertisement


MQA bad for Music?

Reviving this old thread.

I spent some time analysing the analogue output of an Explorer2 while playing 'three-LED' MQA files. Three LEDs lit means that the signal is of 4x rate or higher (i.e. 192k, 176.4k, or DXD rates).

I recorded with my Tascam DVRA1000 running at 192kHz. Theoretically this gives me a view of the MQA DAC's output up to 96kHz. However, the PCM1804 ADC in the Tascam has a large bump of sigma-delta modulator noise above 60kHz, so for payload ultrasonics to be discerned in the 60-96kHz band they, the ultrasonics, would have to be very very loud.

Normal music does not have high-level ultrasonics. In fact most music does not contain much above 30kHz. This makes analysis hard, because in order to spot the effect of upsampling in the signal chain you want to find pairs of matching frequencies symmetrically around Fs/2. Luckily many studios are a bit dirty, EMC-wise, and have spurious signal related to CRTs and switch mode supplies running through their cables. These signals make it onto recordings and then act as tell-take signs...

A-ha:

1eOQi5u.jpg


Spuriae mirror clearly around 48kHz. This is a 96k signal upsampled to 192k.

Eagles:

RLfob8z.jpg


Spuriae mirror clearly around 48kHz. This is a 96k signal upsampled to 192k.

Some jazz track, don't remember what (Bill Evans?)

ZkMCGbE.jpg


Same story. In addition, the master was heavily filtered while converting to 96k.

Roberta Flack:

5YPM4Eo.jpg


Spuriae mirror clearly around 48kHz. This is a 96k signal upsampled to 192k.

Television:

WCuXaoi.jpg


Idem.

2L, possibly the Mozart

v1mrNpR.jpg


Clean recording, no spuriae. All the 2L recordings I tried had very little treble content, too littly to allow searching for images. This was the best. Even so I had to eyeball this in real time (all other graphs were averaged), and freeze the spectral plot when the suspected image finally peaked above the noise bump. You can see that here at 26kHz and 70kHz.


In case you wonder ... all examples, except 2L, contain a strong pair at 29kHz/67kHz. This is not an artefact of my setup. I verified that the 29kHz component is present in the digital signal as it comes from Tidal.
To check that I understand you-all of the files apart from the 2L are upsampled 96kHz with a filter with not much image rejection.
Does this mean that they were originally 96khz i.e. that they have not had the lazy downsampling from 192 to preserve the >48kHz material as aliases which then re-expand to the 48-96khz zone on upsampling (I think this is what Jim calls the origami)
 
I think these all started as 192k, with lazy downsampling to 96k, followed with the origami to 48k (Jim's bitstacking). The original 48-96k information gets annihilated.

But let's remember that the MQA documentation already hinted at this. This is just confirmation. No surprises.
 
The 96k rate -> 192k rate 'origami' could explain why the components mirror as they do. If so, confirms the feeling that the process leaves the artifacts in its wake. However it is curious that all examples except 2L have the same components. Is this something that tends to happen in MQA or that they all (except 2L) had the same spurion (spurious tone) applied before MQA came into play? Or a watermark? Or?...
 
The original 48-96k information gets annihilated.
not quite annihilated (except I suppose in the sense of the poor folk at Hiroshima whose shadows were supposedly burnt on the wall)
Isn't that info preserved as an alias, whose image is then mixed up with the images of 24-48 kHz.

Still that appears to answer the question as to which of the 2 MQA stories about >48khz material is true
 
Aliasing is annihilation, because you can't unmix it afterwards.

This said, I am sure MQA dimensions the first AA filter so that any aliases in the 0-22k band fall below the programme noise.


I'll have a close look at the Hiroshima walls, soon.
 
Aliasing is annihilation, because you can't unmix it afterwards.

This said, I am sure MQA dimensions the first AA filter so that any aliases in the 0-22k band fall below the programme noise.


I'll have a close look at the Hiroshima walls, soon.
Business or vacation ?
 
The 29/67k pairs are curious. I presume the implication is that the 'hf hints' in the bitstacked 48k/24 MQA must generate the 29k? Maybe this is a sign of something like the 'key' sequence in the XORed LSBs? But then, why not in the 2L? Maybe a 'Tidal' watermark?
Their absence in the 2L recording implies that it is not a MQA artifact.
More likely there is an ADC in a processing lab, being used for the recording label digitisations and it has a switched mode power supply nearby.
This would also explain some of the "sameness" rumours if there is one setup doing a lot of the MQA capture and encoding
 
Indeed. MQA generating these tones as part of the system is not very plausible.

On the other hand, all of these recordings being converted with the very same setup also strikes me as not very likely. Time may tell. Not that this is very important.

--

Vacation. Sort of. Friend became dad. Must visit babies. Oh, and we have three sort of famous jazz bars in the vicinity of our Kyoto hotel. This to get back on topic.
 
Indeed. MQA generating these tones as part of the system is not very plausible.

On the other hand, all of these recordings being converted with the very same setup also strikes me as not very likely. Time may tell. Not that this is very important.

IIUC All the examples are 'Tidal' except for the '2L' one. If - as I assume - the 'Tidal' examples *don't* all come from the same label this presumably implies either that:

A) All the 'Tidal' ones were run though the same (non '2L') encoding system which somehow added the tone(s)

or

B) It is a result of something like 'watermarking' by 'Tidal'.

may be the case.

Either way, looks like a nice sign supporting the prediction that the 'folding' method adds aliased copies of HF which decoding can't remove.
 
Disagree. There is no sign of aliasing. All this suggests is that all or nearly-all '192k' MQA is conveyed as 96k followed with in-DAC upsampling with creation of images.
 
Disagree. There is no sign of aliasing. All this suggests is that all or nearly-all '192k' MQA is conveyed as 96k followed with in-DAC upsampling with creation of images.

What does that mean in plain language? :)

Is MQA likely an 'improvement' on plain FLAC files as given to us by Tidal or has it something added/subtracted to/from it?

Is it simply a guarantee that the original is from a Master but doesn't guarantee any better SQ?

Thanks
 
Disagree. There is no sign of aliasing. All this suggests is that all or nearly-all '192k' MQA is conveyed as 96k followed with in-DAC upsampling with creation of images.

I suspect we may be using different terms here.

The 29k / 67k mirror pair seem to me like aliasing/imaging given their symmetry about 48k. What *isn't* clear is which one is the chicken and the other its egg, or if something generates them both.
 
What does that mean in plain language? :)

Is MQA likely an 'improvement' on plain FLAC files as given to us by Tidal or has it something added/subtracted to/from it?

Is it simply a guarantee that the original is from a Master but doesn't guarantee any better SQ?

Thanks

To phrase this in injuneer-speak more plainly:

He means that the signal is at best what you'd get from a 96k sample rate, but with added HF crap.

The examples show HF tones that are *very* unlikely to be from the music, but have been added in some way. And the amount of real HF seems to be minimal beyond the band 96k rate would cover. This last point won't surprise many audio engineers. But of course MQA focus their claims on 'timing' arguments which, erm, could be said to 'blur' this point. :)
 
To phrase this in injuneer-speak more plainly:

He means that the signal is at best what you'd get from a 96k sample rate, but with added HF crap.

The examples show HF tones that are *very* unlikely to be from the music, but have been added in some way. And the amount of real HF seems to be minimal beyond the band 96k rate would cover. This last point won't surprise many audio engineers. But of course MQA focus their claims on 'timing' arguments which, erm, could be said to 'blur' this point. :)

Thanks :)
 
I suspect we may be using different terms here.
The 29k / 67k mirror pair seem to me like aliasing/imaging given their symmetry about 48k. What *isn't* clear is which one is the chicken and the other its egg

Looking at the actual music, dynamically, it is clear that the 29k is the chicken. So the 67k is the imaged egg.
 


advertisement


Back
Top