MQA technical analysis

[Miska]

Though very few speakers (headphones?) will reproduce most of that range. But it probably isn't a good solution to create problems further up the chain.

 

You can always extend the range using additional supertweeters.

 

There is another school of thought that says this is exactly the problem with hi-res: 44.1 captures all the audible effects of intermodulation with ultrasonics that occurred during the recording, so using higher rates is just asking for trouble.

 

I let those people have their RedBook. I still want my hires, preferably in DSD256 format. Thank you.

 

If extending range as little as to 100 kHz causes trouble, then people really should look for better gear instead of blaming it on too good source material. You can always have additional low-pass filter in your playback chain if you want to, but you cannot remove one that has been applied before.

 

If you put a filter close to 20 kHz or so, it will have sonic implications. Even 100 kHz is relatively close. 1 MHz is a good point. (I'm roughly in line with Spectral Audio on that)

[Miska]

I suspect, you refer to noise floor -120 dB (modern DAC).

 

I'm not talking about noise floor of DAC's, those are unimportant for the subject. What matters is noise floor of the recording, which includes acoustic background noise of the recording space, noise of the microphone preamps and the ADC. One can make recording at high enough sampling rate and then determine and decide sufficient distribution sampling rate at the mastering stage based on analysis of the final master. So what I was looking at is the original 2L's DXD master files.

 

It allow to see (don’t hear) not only 56...90 kHz harmonics, but 150 kHz too. It is not real figures, of course, but suggested as example only.

 

Yeah, 150 kHz is not a problem at all for DSD512 which can be pushed to have noise floor well below thermal noise for the 200 kHz bandwidth. DSD1024 is coming soon (and already used by some DACs).

 

So for the Sanken mic I was talking about, using the new ADCs running at 705.6/768 kHz PCM could be a good choice. Or why not DSD256.

[Miska]

I'm agree. DSD1024 is nice thing for transferring 100 kHz band. May be more too, but I don't checked yet.

 

Well, it allows 400 kHz wide band with noise floor below thermal noise. For 100 kHz, DSD256 is enough.

[Miska]

Would you be able to post a pic of that? Just so it's easier to visualize. What about the flac version as well output from the same dac to see if it matches the theoretical?

 

If you look earlier in this thread, I posted some results.

 

By the way, what is the waveform measuring? Some noise level at a specific instant of the recording? It looks like near silence with the highest signal being -60db so it surprises me it doesn't match with the close correlation archimago's blog shows after decoding.

 

Same noise is there throughout regardless of the position.

[Miska]

Does anyone have a definitive answer for this? If it can provide just the render function, I have some ideas for experiments.

 

I'll try to check this on Monday or Tuesday. I need to send the digital capture of the software decoded file because Tidal application enforces hardware decoding when it detects known MQA DAC.

[Miska]

If you mean the spectrograms, I unfortunately still haven't figured out how to interpret them to draw any conclusions and especially not how to compare those with the graphs mansr has provided. Would you be able to explain?

 

I thought it is apparent. Same information, but with one extra dimension - time. The Y-axis of mansr's graphics are color-coded and the frequency is swapped to the Y-axis because time is on X-axis.

[Miska]

I won't stop what I'm doing, but if Chris desires, I have no choice but to talk less about it here.

 

Just let us know if you take the discussion elsewhere...

 

AFAIK, reverse engineering is legal in most places for interoperability purposes:

https://en.wikipedia.org/wiki/Reverse_engineering#Legality

[Miska]

The result is mathematically a different signal compared to the original PCM master. But it may be a better sounding signal, or even a more "accurate" representation of the original sound.

 

I guess you could say that MQA tries to anticipate the damage caused by the ADC and DAC and then applies DSP at various stages in the digital chain in order to remove that noise and reproduce the original analog-like sound.

 

Does it work? We have to decide ourselves. That's the consumer's task So far, Tidal Masters are sounding rather nice.

 

I would rather take an objective view of the performance, rather than subjective.

[Miska]

Thanks, much appreciated.

 

I just checked and the decoded MQA stream from Tidal doesn't trigger the render mode (noise) on the Meridian DAC... So it is played just like plain normal hires file.

[Miska]

Is the Explorer 2 running latest firmware?

 

Yes it is...

[Miska]

If you are looking for good source recorded by DxD and with various converted DSD/Wave/MQA formats

http://www.2l.no/hires/

 

As:

- Finzi: Come Away, Death Marianne Beate Kielland, mezzo soprano Sergej Osadchuk, piano

- Bøhren/Åserud: Blågutten HOFF ensemble

- Groven: Undring Sigmund Groven, harmonica Iver Kleive, organ....

 

Unfortunately not many contain notable amount of high frequency content that would really push MQA into noise modulation (running short on HF encoding bandwidth/stealing bits from lower frequencies). For that you want synth music or rock/blues/pop with close mic'ed crash cymbals and such.

[Miska]

It shows that the folded part also contains information below Fs/2, possibly suggesting something about the filters used in the lossless spectral split and join at Fs/2.

 

You mean lossy split?

 

It is easy to make the split much better without overlap/aliasing around Fs/2...

[Miska]

I suppose, the integration can't decrease level noise to 40 dB.

 

What is level (in dB) of the signal (by oscillogramm) in LFSU scale?

 

Level of random noise is directly dependent on length of the FFT (as well as the other parameters). More frequency bins you have, lower the level of random noise is per bin. Level is constant, you just choose how many bins you distribute it to.

[Miska]

Exactly, and that's why you typically normalise the values so outputs of FFTs of different lengths can be readily compared.

 

Hmmh, discrete tones give pretty much same levels on all cases, but random not, so what do you normalize?

[Miska]

MQA seems sensitive to ID3 tag contents. I took an MQA wav file, and converted it to flac in two flavours: one with ReplayGain writing to the ID3 tag, one without.

 

The Explorer2 refuses to recognise the ReplayGained track as MQA, even though the actual signal in the file nulls perfectly with the original MQA wav.

The Explorer2 is fine with the non-RGed track.

 

Remember that one purpose of MQA is to allow decoded playback only in situations and environments tightly controlled by MQA. So you cannot apply any DSP of your choice; like replay gain, digital room correction, headphone cross-feed or stuff like that, unless those are specifically approved/blessed by the MQA company.

[Miska]

We've all seen the characteristic noise hump in undecoded MQA files. Apparently, this is the result of shaped dither applied during the encoding. The first two steps of the decoding process entail removing part of this noise. This is possible since the pseudo-random sequence used to generate it is known. Here's a graph of the decoder in action

 

What I've understood, it is not actually shaped dither, but the encoded upper band data that has been noise-shaped to be less audible and then mixed with the actual shaped dither (that is not removed).

 

So how I see the graph is that the difference is the actual useful data taken out.

 

Of course I'm not at all sure if I'm right...

[Miska]

1. Hardware decoded files show signal information above 48khz which the software decoded files don't.

 

Where? So far what I've looked at for example 2L content compared to the original 352.8 kHz DXD and proper 192k down conversion of those, the MQA content rolls of early (starting at about 30 kHz) compared to the original or 192k versions where content reaches up to about 60 kHz.

 

If it was simply upsampling 96khz I would expect a sharp cutoff at 48khz and nothing above. So his doesn't sound like upsampling

 

MQA filters are specifically extremely non-sharp and "poor", so they leak a lot.

[Miska]

The low 8 bits of a 24-bit MQA file contain a compressed representation of the 24-48 kHz frequency content. The decoder combines this with the base band information in the high 15 bits to recreate a 96 kHz stream. This part works more or less the way they say it does.

 

What I've been wondering is what happens when MQA encoder is running short in bits to encode the upper band. From what I've seen I get feeling that it may be stealing bits from the lower band...

 

If one could encode 0 dBFS 192/24 white noise and multi-tone, we could know more. But so far the best I have are the synth tracks that actually push stuff out throughout the bandwidth of 96/24 original.

 

For example MP3 encoders look OK with single sine, but go pretty bad at multi-tone and white noise (of course).

[Miska]

When I'm talking about sound information above 48khz, I'm not just talking about mirror image aliasing. From some of the graphs archimago blog shows, it matches the original hires content pretty closely up to 60khz.

 

The resolution is not so great on the spectrograms, but what I see by quick glance is just images leaking through the filter.

[Miska]

Given that Bob Stuart and Peter Craven have some ability to understand digital filters, can you make a guess as to why (if true) they elected to implement filters that are "poor and leak a lot."

 

Because it's the "non-ringing" school of thought, same category as NOS DACs and for example Ayre's "listen" filter. You can check out the Stereophile measurements linked here to see... This is complete opposite of for example Chord which goes to the other extreme. I personally rather stay somewhere in the middle, but for example HQPlayer offers also both extremes as an option. DAC chips have long had options for different schools too, with usually two filter choices.

 

In these measurements, there's a 19.1k tone at 44.1k sampling rate and there shouldn't be much if anything of it left at 25 kHz (blue). A leaky filter is one that leaves strong tone visible at 25k. This is image of the 19.1k tone. Red plot is just white noise at 44.1k sampling rate, thus spanning up to 22.05 kHz, you can see how much of it is leaked above 22.05k frequency as images.

 

Mytek Brooklyn's MQA-filter:

 

Meridian Explorer2:

 

See a striking similarity between the two?

 

Ayre Codex:

 

For comparison, Chord Mojo:

[Miska]

Are they poorer in this regard than any (meaning all) of Meridian's non-proprietary filters?

 

Sorry but I don't understand the question, what do you mean by non-proprietary?

 

The earlier Explorer (non-2) was better in that respect:

[ATTACH=CONFIG]33144[/ATTACH]

 

Or if you look at the original Prime:

[ATTACH=CONFIG]33145[/ATTACH]

(possibly changed later as part of MQA firmware update?)

[Miska]

Miska, your images are showing as non-valid attachments.

 

Crap, the forum acts up sometimes with images, first shows those and then they silently disappear and become broken...

 

Explorer:

 

Prime:

[Miska]

Input signal is 19.1 kHz pure sine in 44 kHz?

 

Yes, just normal 19.1k sine at 44.1k sampling rate.

 

It is modelled signal? What is bit depth?

 

What you mean by modelled signal?

 

Should be at least 24-bit, but probably 32-bit. I use dithered 32-bit test signals, and sometimes just 64-bit float WAVs.

[Miska]

As "modelled" I meant synthesed sine into audio file.

 

Yes, that's of course the case.

 

How you send direct to DAC 32- or 64-bit float? A DAC can receive integer formats only, as far as I know.

 

It is automatically dithered by the player to DAC's resolution what ever that is, or converted to DSD when that is the output format.

 

It is practical for many cases, because there so many different DAC resolutions, 16, 18, 20, 24 and 32 bit.

[Miska]

It's not that simple. Even you could just go out and buy one, which you can't, the terms of use would most likely explicitly prohibit the publication of anything resembling test results. They wouldn't be the first do that kind of thing.

 

But if you get someone to encode test signals for you and then you publish the results...