MQA

It makes sense that non-oversampling stuff w/o any digital filtering may indeed benefit from higher sampling rates the most due to less roll off, with perhaps some noise and distortion issues depending.

I heard one of the Metrum (I think) NOS DACs, and indeed it seemed to sound a little relatively rolled off in some cases, if I remember correctly.

The digital FM channels may experience some lossy signal processing, which may cause performance degradation.

 

Wow, no error correction?! From what I'm reading, on top of that, the digital channel is transmitter at much lower SNR.

If they did their link budget right they may end up compressing the crap of the signal to keep the transmission reliability up. That's not awesome.

 

Just read an SBR paper for HD-Radio (http://www.stockis.se/AES112th_SBR.pdf). It sure sounds a lot like that MQA stuff.

It seems the do include some sort of CRC and error "resilience" (correction?) stuff in it.

In SBR (HD-Radio) they basically do not transmit the high frequencies and instead recreate them based on the low frequencies along with some high frequency envelope information, and high frequency signal parameters. They use the "psychoacoustics" model to make sure that crappy stuff due to the reconstruction will be masked as much as possible. Now, what do they cut off? About half the signal BW (perhaps 10 kHz?). In the paper example, all things above 6 kHz where castrated, and reconstruction was done to about 15 kHz.

BTW, the SBR signal gets further compressed...

Again, this does sound a bit like MQA from the rough sketches.

Note all these approaches seem proprietary, patented or whatever. So no details are given on the parameters extracted to aid high frequency component estimation. Just some rough ideas here and there.

The aim here is not signal quality, but superior compression with "acceptable" performance. This is not high definition IMO, and this may be why The Merv No-Like HD-Radio.

The apparent difference between SBR and MQA seems to be that the stations pay for SBR arguably to make more efficient use of their allocated BW and to perhaps lower transmission power to save on the electric bill. With MQA (if MQA is like SBR), the consumer pays for a compressed "acceptable" performance IMO with no apparent gains (it may not lower ones electric bill and the consumer proly could care less about allocated BW).

EDIT: Here is a bit more info on SBR (HD-Radio) and does cover some of the details about estimation:
http://www.db-thueringen.de/servlets/DerivateServlet/Derivate-29663/ilm1-2014200076.pdf
Again, this is not something I have worked on or stuff. But it seems SBR is aimed at "acceptable" perceived performance with high efficiency (low bit rate), not high definition or high fidelity or so forth.

 

LOL! yes, many things don't even error correct. It is possible they rely on other error control mechanisms at higher layers though.

In any case, many folks have voiced their concerns about MQA:

1) http://www.superbestaudiofriends.org/index.php?threads/mqa.1635/#post-54411

2) http://www.superbestaudiofriends.org/index.php?threads/mqa.1635/#post-54508

3) http://www.superbestaudiofriends.org/index.php?threads/mqa.1635/page-2#post-54521

4) http://www.superbestaudiofriends.org/index.php?threads/mqa.1635/page-2#post-54573

5) http://www.superbestaudiofriends.org/index.php?threads/mqa.1635/page-2#post-54589

Merv mentioned HD-Radio as well. The similarities I see with MQA is that MQA also sacrifices high frequency fidelity for efficiency. It perhaps does not loose as much as SBR (Spectral Band Replication) in HD-Radio, but it's still a lossy approach.

As many here, I don't see MQA going anywhere.

 

Yes, they are different. Here is Bob's paper, probably more closely related to MQA: http://www.aes.org/tmpFiles/elib/20160529/17501.pdf.

But both deal with compression of a signal for convenient delivery due to BW limitations. As Bob's puts it:

"...we should always store the best representation and it will rarely make sense to distribute these core assets in a format identical to that of the archive."

I feel the meat in Bob's paper starts in Section 4 (most of the previous introductory sections are very fragmented IMO). As far as Section 4, I have lots of issues with his "Sampling" sub-section. His discussion about the impulse, frequency-domain relationships and pre/post ringing are grossly way off the mark. So much that I don't care for the rest of the paper.

@Azteca, MQA as far as I can tell is partly (and perhaps mostly) a compression algorithm for efficient delivery of content. In that sense it resembles HD Radio (though arguably not as extreme... maybe). It is possible that BW constraints are not as extreme as with HD Radio, but when delivering streaming content, they do exist.

It is not a lossless, archive quality content delivery scheme. And I fail to see clearly what it brings to the table vs true lossless or even well implemented lossy formats that are license fee free or close too. Or more importantly, why there is a need for a scheme that imposes HW restrictions to the A/D and D/A converters. Bob's paper IMO fails to convincingly address the need for these HW restrictions (too many issues in the paper's logic).

 

Compression is important in both audio and video (or any information conveying system). However, my quick brush on MQA doesn't seem to be delivering the punch line of the story. I don't see it's need, and the technical discussion seems broken to me.

 

Darko's earlier reviews were really good. Then he sort of hit a spot where everything was good. He seems a little bit better now.

 

For DACs, it's not aliasing one should be concerned about, unless one decimates. Usually the signal is already bandlimited. It is images one is concerned about at this point. There are many ways one can think of something like a MegaBurrritoCombo. An interpolator, or a anti-imaging filter. One gets better than 12 kHz there. There is also an analog filter stage to get rid of images past the Nyquist of course (you can think of it as yet another interpolator).

For recording and ADCs, 96 kHz indeed helps, but even if one uses 48 kHz, I wouldn't call 12 kHz the frequency threshold at which things start to go bad. It depends on the analog filter and so forth.

Note the number of bits has little to do with sampling rate. I could sample something at 1 MHz, with 6-bits of resolution. Unless I apply some sort of quantization noise shaping to deal with the relative small number of bits, such an approach will probably do poorly for audio applications.

18-20/96 PCM is also not "perfect". The rate and bit width is IMO good for audio applications.

Note the recording may have more or less effective bits than 18-20 bits.

 

I don't give a rats ass about theory. You guys can masturbate to theory all you want. Also, make sure you understand your theory right. There is a lot of misinformation out there. All people need to know is that 44.1kHz is sufficient to perfectly reconstruct signals up to 20kHz assuming infinite bit-depth.

Math is magickal. Remember that line in the Thor movie: what is magic to you is science (math) to people who understand it. Yeah, that's kind of how it works. So mere mortals are probably better off not trying to make sense of how Thor's Hammer or digital reconstruction works. I mean, you don't see me trying to understand how Boom Tubes or Motherboxes work in the DC universe. If I did, Darkseid would just laugh at me and call me stupid.

Finally, all I can say is starting from a DSD128 source and down-converting to various formats, sampling rate more than 44.1kHz made zero difference, but 18 to 16 bits did make a small difference to my ears.