MQA spectrum plots

Using an MQA decoder extracted from Bluesound Node 2 firmware image, I have managed to get some real data to discuss. For a first look, I picked the 2L-064 sample from the 2L test file collection.

The spectrogram of the DXD original looks like this (left channel only):

To better see the music content, here's the same file reduced to 88.2 kHz:

As we can see, there is content approaching 40 kHz in places.

This is the spectrogram of the MQA file without decoding:

The frequency peaks extend right to the very top, indicative of either a very sharp filter or some amount of aliasing. A band of noise is clearly visible above 16 kHz.

Decoding the MQA file to 88.2 kHz produces this spectrogram:

A few things are obvious here:

1. The noise band is still there.

2. There is much less high-frequency content than in the original file.

3. There's a smattering of artefacts at the top of the frequency range.

Looking at the average spectrum for the entire file, we get this graph:

Here we can see that the decoded version actually has somewhat less of a hump around 20 kHz than the undecoded MQA. Above 22 kHz, the decoded MQA falls off much quicker than the original, and there is clearly something funny going on above 40 kHz.

Finally, we look at the silence at the start of the track (note different range):

The MQA file, both before and after decoding, has a somewhat higher noise level already at 1 kHz, and around 6 kHz the curves really start diverging. Again, decoded version deviates from undecoded MQA above 20 kHz. Around 26 kHz, the decoded MQA drops below the noise floor of the original. Evidently, some kind of filtering has taken place though its exact nature is impossible to determine from this graph alone.

So, I am not very wrong if I wrote in other thread:

In reality MQA cut's down bits and separates useful signal range to HF and LF part, where LF part carries useful compressed audio and HF part carries DRM codes and some higher audio signals if they are present. Priority is DRM, not higher audio signals because DRM is a must.

I think, MQA is clever lossy delivery container, where some additional data is encoded directly to higher octave audio - that's why in case of streaming we can play stream without headers.

Lossy because spectrum clearly differs from original, significantly.

Great stuff @mansr ! I was planning to do the same, but didn't find time. But I did quite a bunch of tests through a Meridian DAC with MQA decoding to compare the performance from analog domain and got similar results.

My overall view has been that in order to reduce the higher octave (>22.05/24k) data enough to fit the encoding band in lower octave LSBs, they run slow roll-off filter that cuts at about 30 kHz. This also matches with their filter kernel discussion. When there's not much content to encode, encoding the rest is not so much problem and since it's not wide, aliasing/folding it down around the 22.05/24k can be masked with the shaped noise that bumps at the top of the lower octave. (the matching MQA reconstruction filter that can be seen at the Stereophile's measurement of the Mytek Brooklyn DAC also follows the same pattern)

I sort fail to see the point they advertise that the MQA encoded data can be unfolded to 4x and 8x rates, because there's really nothing left to unfold anymore...

What I find funny is that they emphasize a lot about transient performance, yet they severely cut on the high harmonics needed to represent fast transients, actually slowing down the transients.

@mansr if you have time, it would be interesting to see difference spectrums of the MQA data compared to the original, something like I did long ago here to compare various lossy codecs. The process was to first time-align the two files and then calculate average spectrums of source, target and the difference tone. And these days when I know more I would also plot out minimum and maximum difference spectrums (peak-hold lowes/highest value per bin). This would tell something about possible changes in the lower octave too.

This also shows one thing I especially noticed from my tests, there's quite a bit of aliasing evident at the top of the decoded MQA spectrum that isn't there in the original! Closer the amount of energy in the source goes to 44.1k Nyquist of the 88.2k sampling rate, stronger the aliasing there is too. I first thought it is only due to bad oversampling filter in the DAC, but apparently it is really part of MQA itself.

The point with MAQ is, it requires in addition an altered DAC gear what is DAC silicon dependent.

This means, each DAC gear has to send to the holy company for adaptation.

In other words, the final sound will be gear dependent too (in addition to the MAQ treatment as may psycho acoustic alog's as on MP3).

Hp

@mansr - Can you reduce the non MQA file (second image) to 20bit but keep the same 88.2khz sampling and compare the size with the MQA file? Thanks!

Using the highest FLAC compression (level 8), I get these sizes:

MQA: 31M

88.2k, 20-bit: 34M

88.2k, 19-bit: 29M

88.2k, 18-bit: 25M

88.2k, 17-bit: 21M

At 19 bits or higher, the spectrum is practically indistinguishable from the original DXD. 18-bit and lower shows a slightly elevated noise floor in the silent parts.

I think, MQA is clever lossy delivery container, where some additional data is encoded directly to higher octave audio - that's why in case of streaming we can play stream without headers.

Lossy because spectrum clearly differs from original, significantly.

There maximal level about -105 ... -110 dB. It's about CD quality.

If I don't mistaken, CD quality of packed form was promissed.

Looks like it work as promissed (if consider level noise as quality mark).

Though used 24 bit container.

If I don't mistaken, CD quality of packed form was promissed.

CD quality from HiRes files - good achievement :).

CD quality from HiRes files - good achievement :).

I can suppose only, what MQA as CD transport is secondary aim of MQA developers.

As compatibility issue only.

Yuri, pervõi aprel jesjo ne prizol . Serjozno...

@mansr if you have time, it would be interesting to see difference spectrums of the MQA data compared to the original, something like I did long ago here to compare various lossy codecs. The process was to first time-align the two files and then calculate average spectrums of source, target and the difference tone. And these days when I know more I would also plot out minimum and maximum difference spectrums (peak-hold lowes/highest value per bin). This would tell something about possible changes in the lower octave too.

Spectrogram and average/max/min spectrum for the difference between decoded MQA and DXD downsampled to 88.2 kHz:

Using the highest FLAC compression (level 8), I get these sizes:

 

MQA: 31M

88.2k, 20-bit: 34M

88.2k, 19-bit: 29M

88.2k, 18-bit: 25M

88.2k, 17-bit: 21M

 

At 19 bits or higher, the spectrum is practically indistinguishable from the original DXD. 18-bit and lower shows a slightly elevated noise floor in the silent parts.

Interesting... It looks like one would prefer the 20bit version sans all the MQA noise. So in essence, we have MQA, a lossy and noisy process that sacrifices bits for the sake of reducing file size..

CD quality from HiRes files - good achievement :).

LOL (really am)

So in essence, we have MQA, a lossy and noisy process that sacrifices bits for the sake of reducing file size..

In same time file size don't decrease as much it would (compared to FLAC), so this process is ineffective and uses resources to something else, maybe for DRM and so called "origami", where unfortunately some information is lost.

In same time file size don't decrease as much it would (compared to FLAC), so this process is ineffective and uses resources to something else, maybe for DRM and so called "origami", where unfortunately some information is lost.

Supposedly DRM isn't in MQA.

According to Hans' Youtube video (advance to 12:13 minutes in the video) he was advised my Merdian that MQA did not use DRM.

From Jesus at Sonore. Link below.

http://www.computeraudiophile.com/f26-sonore-sponsored/microrendu-and-mqa-31157/#post621587

Supposedly DRM isn't in MQA.

According to Hans' Youtube video (advance to 12:13 minutes in the video) he was advised my Merdian that MQA did not use DRM.

It doesn't at present. Since the format is secret and a standalone decoder isn't made available, they could easily make things much stricter in a future version. Even if that doesn't happen, it's bad enough to be stuck with undecodable files should MQA go out of fashion and no longer be supported in new products.

Anyhow, can we please stick to technical discussions in this thread?

 

Anyhow, can we please stick to technical discussions in this thread?

Yes, please! Show more data!

Supposedly DRM isn't in MQA.

The thing is - we don't know for sure.

From that perspective, if MQA don't contain a DRM at all, then compressing is more ineffective, so "ten points" to Bob .

On the Linn forum we found the following link: http://images.computeraudiophile.com/ca/utimaco_mqa.pdf.

DRM and MQA does not seem to be strangers to one eachother. I hope this helps.

(Hi guys. )

Spectrogram and average/max/min spectrum for the difference between decoded MQA and DXD downsampled to 88.2 kHz:

Thanks! This is pretty interesting. Obviously transients have quite a bit of difference. But it is really curious that there's so much peak difference in the 4k and below band! While as expected it noise-wise circulates around the 20 kHz noise bump. Maybe this is related to the noise modulation I was seeing.

On the Linn forum we found the following link: http://images.computeraudiophile.com/ca/utimaco_mqa.pdf.

 

DRM and MQA does not seem to be strangers to one eachother. I hope this helps.

(Hi guys. )

Does the mere presence of cryptographic functions guarantee the presence of DRM?

Does the mere presence of cryptographic functions guarantee the presence of DRM?

Please, take this topic to some other thread.

Thanks! This is pretty interesting. Obviously transients have quite a bit of difference. But it is really curious that there's so much peak difference in the 4k and below band! While as expected it noise-wise circulates around the 20 kHz noise bump. Maybe this is related to the noise modulation I was seeing.

That difference in energy is already visible in the (352.8) original ...

Btw, there's more energey again at ~6KHz and above.

So ask 2L or the musicians.

That difference in energy is already visible in the (352.8) original ...

Btw, there's more energey again at ~6KHz and above.

 

So ask 2L or the musicians.

Difference against what?

Difference against what?

In what Miska thought to see. I saw it as well, but since it's in the original as well it should be normal. Or not MQA related.

The confusion could be in that I looked at the spectral plots to see/notice this and that you thus can see it in the original (352.8) as well, with your downsampled 88.2 to hint at it (in your first post).