Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Meridian Audio's new... sub-format called MQA. (Read 147588 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Meridian Audio's new... sub-format called MQA.

Reply #150 –
Quote from: xnor on
I assumed that tracks were recorded normally e.g. into 24/192, edited, mixed ... and finally the high frequency content is attenuated and aliased into e.g. 48 kHz sampling rate. Isn't that basically it?


IIRC they downsample from 192 to 96 with a very shallow filter (very wide transition band, very narrow impulse response), even allowing some aliasing above 30kHz or so.

Then they take this 96k recording and fold the 24-48k band into the lower bits of the baseband, obtaining a 48k encoded recording with '96k bandwidth' and '192k temporal accuracy'.

Quote from: Porcus on
Not really, if the buzz is to be believed (which is certainly some of an "if"). Seems that they want to reverse ADC artifacts from recordings that were certainly not recorded into 24/192.


This only makes sense for CD-rate recordings where at least two major classes of recording-time anti-aliasing filters can be distinguished:
-PCM1630-style analogue (minimum phase, starting at 20kHz, and -surprisingly- quite a lot of attenuation at Nyquist)
-half-band FIR (linear phase, -6dB or so at Nyquist)

Presumably Meridian would want to replay the former with a half-band linear-phase FIR, leaving the result dominated by the PCM1630, and the latter by a minimum phase apodising filter, leaving the result dominated by the MQA filter.


Meridian Audio's new... sub-format called MQA.

Reply #151 –
Quote from: xnor on
One more thing Porcus, regarding (re)mastering. If you've recorded at 48 or even 44.1 kHz then you cannot create a genuine "hi-res" track from it. MQA doesn't help here either.

I think what they want to achieve is some minimal recording format (say 24/96 or even 24/192) with MQA'd tracks, so that the consumer doesn't have to worry about fake "hi-res". But from a purely technical pov, I don't think anything prevents one from creating a fake "hi-res" MQA file...



Some would argue 16/44.1 is hi-res...

Meridian Audio's new... sub-format called MQA.

Reply #152 –
Quote from: Porcus on
Not really, if the buzz is to be believed (which is certainly some of an "if"). Seems that they want to reverse ADC artifacts from recordings that were certainly not recorded into 24/192.
http://www.audiostream.com/content/tidal-s...mqa-music-files

[...]

Not a "genuine", but if we for the sake of the argument buy into the claim that anti-aliasing filters with artifacts way down in the audible range were employed upon digitizing, then you can certainly try to improve. As you point out, you cannot (re-) create a "genuine" lossless version of what was filtered off, but you can in principle use knowledge of the filter to improve.

(Again, I do not want to sound like I endorse their claims, but there could be more to it than ultrasound mumbojumbo.)

I'm really unsure about this.
On the reversal of ADC artifacts:
As I said, one thing could be handling the high frequency noise produced by ADCs that would only be a waste of space. Especially to PCM converted DSD comes to mind.
Another thing could be that they seem to dislike steep and linear-phase low pass filters, so they could use a slow filter with a lower cutoff frequency to reduce evil ringing? to theoretically get nice looking impulse responses.

But we'd need to hear a technical person talk about what actually happens instead of this marketing stuff.


I don't think that anything changes about digitizing. MQA just seems to compress a high sampling rate PCM file to a low sampling rate one by aliasing the high frequency content and storing it in the "noise bits". It's not lossless if you consider the "noise bits" information.
This is different from aliasing happening during A/D conversion, e.g. a small portion (until the filter reaches its stopband) of signal energy above 96 kHz (given 192 kHz sampling) "folding back". This is usually just pure noise and MQA would have to go to 384 kHz anyway to reproduce it, so I don't see that happening or any good reason for it. You also cannot "unfold" this. Information is lost during sampling.

I could be mistaken though.
"I hear it when I see it."

Meridian Audio's new... sub-format called MQA.

Reply #153 –
Quote from: Porcus on
Quote from: xnor on
One more thing Porcus, regarding (re)mastering. If you've recorded at 48 or even 44.1 kHz then you cannot create a genuine "hi-res" track from it. MQA doesn't help here either.

Not a "genuine", but if we for the sake of the argument buy into the claim that anti-aliasing filters with artifacts way down in the audible range were employed upon digitizing, then you can certainly try to improve. As you point out, you cannot (re-) create a "genuine" lossless version of what was filtered off, but you can in principle use knowledge of the filter to improve.

Try...and fail.

Even if you knew the exact ADC used and the response of its anti-aliasing filter, you simply cannot reliably recreate frequency content above Nyquist.

As an example, suppose you have content sampled at 48kHz and there is a tone at 22k.  How do you know this tone was originally 26k and not simply 22k all along?  What if there were both 22k and 26k in the source, how would you know how much of which?

There is no "in principle" unless you know some law that gets you around Shannon.

Meridian Audio's new... sub-format called MQA.

Reply #154 –
The way I understand it, they convey extra information in a data channel that is hidden in the noise. That's an old trick that people have found many applications for, for example surround information, watermarking, etc.

It is not a great stretch of imagination to use such a data channel for conveying high frequency information. You are not bound by Nyquist here in its strict sense, because the data must first be decoded and processed before it can be used to modify the sound output. It sort of requires that the levels you want to encode in this extra channel are rather low, otherwise you need too much extra data, but I understand that this is a prerequisite that Stuart et.al. have mentioned themselves.

If you want a simple to understand graphical picture, you'd imagine a sound wave at a fundamental frequency, with some superimposed small ripples at higher frequencies, which would go lost when sampled at standard rates. Rather than sampling the entire thing at higher sampling rates, you could sample it at the standard rate and note the difference to the normal sample at several instants before you take the next sample. Since the difference will be small, it can be encoded with only a few bits, rather than spending an entire wordlength.

To put it in another way, it would be oversampling where the in-between-samples are kept and encoded with an efficient difference code. I hope you can understand my attempt at describing it without an actual picture.

I'm not saying that this is what they are doing. It is just meant as a relatively simple example of a sampling method that can capture small amounts of high frequencies while sampling at the base rate. The effect of the entire thing is a kind of tradeoff that surrenders some dynamic range for a small amount of high frequency. This can be made backwards compatible in that the extra data (and with it the high frequencies) are lost when playing it back with a normal DAC that is unaware of the scheme. The lost dynamic range remains lost, however.

It does provide the possibilty to sell (limited) high resolution, in the sense that higher frequencies are contained with restricted levels, at the expense of some dynamic range, while at the same time being compatible with the CD format. In other words, it offers the content providers the opportunity to make a single product that is both "high-res" and can be played back with CD-format equipment, instead of having to produce the product in two separate formats. That could be attractive to some, but it basically kills the possibility to do separate masterings for the two quality levels. And the normal customer with standard playback equipment loses some dynamic range, so the CD-format quality will suffer (again).

Personally, I would use such a data channel for supporting additional headroom instead of higher frequencies. I have no clue whether that can be pulled of in a satisfactory way, but if it can, it would tackle a much more important problem than including high frequency content that a few deluded people believe they can hear.

Meridian Audio's new... sub-format called MQA.

Reply #155 –
Quote from: pelmazo on
You are not bound by Nyquist here in its strict sense, because the data must first be decoded and processed before it can be used to modify the sound output.

As it relates to Porcus's suggestion that you can somehow improve a non-mqa-encoded digitization that is plagued by aliasing, you are bound by Nyquist of the original sample rate.

Even if a down-sampled version is mqa-encoded, you are still bound by Nyquist of the sample rate of the source that was encoded.

Quote from: pelmazo on
Personally, I would use such a data channel for supporting additional headroom instead of higher frequencies.

Something like HDCD, right?  Well, I hope we don't need to have another discussion over the woeful* inadequacy of 16 bits to deliver legitimate content.  I'll also point back to Kohlrabi's last post.

(*) I'm being sarcastic.

Meridian Audio's new... sub-format called MQA.

Reply #156 –
Why on earth do people talk about lossless compression then? It's not.

And why not just take noise-shaped 16 bit, 192 kHz with some real lossless codec to get the same "temporal resolution" and possibly higher dynamic range? That is actually lossless, doesn't cost license fees, doesn't require special hardware and has about the bitrate of CD audio.
For lower bitrates streaming you could add lossyWAV processing.
"I hear it when I see it."

Meridian Audio's new... sub-format called MQA.

Reply #157 –
Quote from: greynol on
Even if you knew the exact ADC used and the response of its anti-aliasing filter, you simply cannot reliably recreate frequency content above Nyquist.


Do they claim only to work above Nyquist?

Quote from: greynol on
As an example, suppose you have content sampled at 48kHz and there is a tone at 22k.  How do you know this tone was originally 26k and not simply 22k all along?


I would suppose a good vinyl pop removal algorithm could by and large improve, despite not possessing the 100 percent sure knowledge of what is noise and what was intended. Wouldn't you?

Meridian Audio's new... sub-format called MQA.

Reply #158 –
Quote from: xnor on
Why on earth do people talk about lossless compression then?

Which people, the shysters of the audiophile world?

Meridian Audio's new... sub-format called MQA.

Reply #159 –
Quote from: Porcus on
Quote from: greynol on
Even if you knew the exact ADC used and the response of its anti-aliasing filter, you simply cannot reliably recreate frequency content above Nyquist.
Do they claim only to work above Nyquist?

I assume you mean Meridian or those who are speculating about mqa.  Either way, I don't know.  Do they?

Quote from: Porcus on
I would suppose a good vinyl pop removal algorithm could by and large improve, despite not possessing the 100 percent sure knowledge of what is noise and what was intended. Wouldn't you?

All sorts of processing can be used to make subjective improvements, but I thought you were talking about removing aliasing or restoring aliased information, in which case you are only guessing about information that has been permanently lost.  If you try to apply some filter using what you know about the ADC, you may get an audible difference and that difference may result in a subjective improvement for one signal, but the same filter may also result in a subjective worsening for another signal.

Meridian Audio's new... sub-format called MQA.

Reply #160 –
Quote from: greynol on
As it relates to Porcus's suggestion that you can somehow improve a non-mqa-encoded digitization that is plagued by aliasing, you are bound by Nyquist of the original sample rate.

Even if a down-sampled version is mqa-encoded, you are still bound by Nyquist of the sample rate of the source that was encoded.

Well, sure. The MQA encoding requires compatible gear at both ends to pull off the trick. You can't magically after-guess spectral content that was lost in sampling. There are techniques which try that, but I consider them a fraud.

Quote
Something like HDCD, right?  Well, I hope we don't need to have another discussion over the woeful* inadequacy of 16 bits to deliver legitimate content.  I'll also point back to Kohlrabi's last post.

(*) I'm being sarcastic.

HDCD doesn't seem to to do much on the headroom side; it improves the noise floor (but I admit that I don't remember much about it).

I share your sarcasm here, because there's an implicit and tacit admission in the MQA scheme: That 16-bit audio has enough dynamic range to sacrifice some of it for other purposes. I wonder how much of that has registered with the audiophiles who are praising MQA.

Meridian Audio's new... sub-format called MQA.

Reply #161 –
Quote from: xnor on
And why not just take noise-shaped 16 bit, 192 kHz with some real lossless codec to get the same "temporal resolution" and possibly higher dynamic range? That is actually lossless, doesn't cost license fees, doesn't require special hardware and has about the bitrate of CD audio.

That's not backward compatible with the CD. I think that's the major point here which makes their scheme attractive to content providers who would have to offer a CD-compatible version anyway.

Meridian Audio's new... sub-format called MQA.

Reply #163 –
Quote from: pelmazo on
That's not backward compatible with the CD. I think that's the major point here which makes their scheme attractive to content providers who would have to offer a CD-compatible version anyway.

Ironic, isn't it?

Meridian Audio's new... sub-format called MQA.

Reply #164 –
Quote from: pelmazo on
HDCD doesn't seem to to do much on the headroom side; it improves the noise floor (but I admit that I don't remember much about it).

IIRC, peak extension provides an increase in dynamic range at the expense of a little noise.

Meridian Audio's new... sub-format called MQA.

Reply #165 –
Quote from: pelmazo on
That's not backward compatible with the CD. I think that's the major point here which makes their scheme attractive to content providers who would have to offer a CD-compatible version anyway.

The way I understood it the main problem is the doubling of bitrate with doubling of sampling rate. Streaming actually seems to be their main point, which requires decoding software anyway..
Also, 24/48 (which seems to be the preferred format) is not actually CD-compatible. Even with 44.1 kHz you have to do a conversion.

And lastly, you can always offer resampled CD versions. That is how it's done actually for quite some time now. If you additionally compressed that losslessly, you will get even lower bitrate.
"I hear it when I see it."

Meridian Audio's new... sub-format called MQA.

Reply #166 –
Yes, but you might lose fidelity that can potentially be detected 56% of the time with cherry-picked hardware and also assuming you are using a "typical" digital filter.

Meridian Audio's new... sub-format called MQA.

Reply #167 –
No no, with 16/192 and proper lossless compression you wouldn't lose anything. Actually I think you'd achieve higher quality than MQA but without a proprietary format or license fees.


As for the argument "If the receiver cannot decode the algorithm, it will be just played as a CD-quality stream": in our day and age a receiver is a computer, smartphone or portable audio player. Most of them can decode something like FLAC.
And nobody, except some crazy audiophiles, stores music in an uncompressed PCM format nor is music streamed in such a format.
"I hear it when I see it."

Meridian Audio's new... sub-format called MQA.

Reply #168 –
Quote from: greynol on
Ironic, isn't it?

I'd call it devious.

Lets for a moment assume the stance of a content provider who doesn't care about quality, but rather about his revenues. What would his interests be? I'd say:
  • He wants copy control (he'll call it "rights management", because it sounds rather more harmless)
  • He wants compatibility with the CD format, because that's still the most widespread; a sort of least common denominator that suits the largest number of people.
  • He doesn't want several different formats, because that causes cost and problems.
  • He realizes that he can't copy control the CD-format; that's been tried and it failed. Repeatedly.
  • Hence he's prepared to surrender copy control of a limited-quality version of his product.
  • As this is unavoidable, he's going to want the lowest quality the market tolerates for the CD-format "common denominator".
  • The native CD format is much too good for that, see the success of the much inferior LP.
  • That means that the CD has spare data capacity, which might be used intelligently.
  • So why not limit the baseline CD to the performance of the LP, mainly by raising the noise floor some 20 dB (or more if it is spectrally shaped), and convey extra data in the noise?
  • Now the question is how to intelligently use the extra data channel. Copy control is a given, but that requires only a small fraction of the capacity.
  • The question is how to use the spare capacity to convey "extra quality" the consumer is likely to crave for hard enough to swallow the bait.

Meridian Audio's new... sub-format called MQA.

Reply #169 –
Quote from: xnor on
No no, with 16/192 and proper lossless compression you wouldn't lose anything. Actually I think you'd achieve higher quality than MQA but without a proprietary format or license fees.

Tongue in cheek.

I can claim improvements all day long if you aren't allowed to hold me down to something specific.

Anyway, I was answering the second portion of your post:
Quote from: xnor on
And lastly, you can always offer resampled CD versions. That is how it's done actually for quite some time now. If you additionally compressed that losslessly, you will get even lower bitrate.

Sorry that I didn't quote it.

Meridian Audio's new... sub-format called MQA.

Reply #170 –
Quote from: pelmazo on
  • The question is how to use the spare capacity to convey "extra quality" the consumer is likely to crave for hard enough to swallow the bait.
[/li][/list]...for hf content no one will ever hear at the expense of noise which someone might very well be able to hear.


Meridian Audio's new... sub-format called MQA.

Reply #171 –
Quote from: greynol on
...for hf content no one will ever hear at the expense of noise which someone might very well be able to hear.

That's the problem. Is it going to be enough when they can be made to believe they hear it?
 

Meridian Audio's new... sub-format called MQA.

Reply #172 –
Quote from: greynol on
Quote from: Porcus on
Quote from: greynol on
Even if you knew the exact ADC used and the response of its anti-aliasing filter, you simply cannot reliably recreate frequency content above Nyquist.
Do they claim only to work above Nyquist?

I assume you mean Meridian or those who are speculating about mqa.  Either way, I don't know.  Do they?


I do not know either, I asked because you used the "above" word. Myself I just hadn't even imagined they would express a reservation that it only has effect above Nyquist.


Quote from: greynol on
Quote from: Porcus on
I would suppose a good vinyl pop removal algorithm could by and large improve, despite not possessing the 100 percent sure knowledge of what is noise and what was intended. Wouldn't you?

All sorts of processing can be used to make subjective improvements, but I thought you were talking about removing aliasing or restoring aliased information, in which case you are only guessing about information that has been permanently lost.  If you try to apply some filter using what you know about the ADC, you may get an audible difference and that difference may result in a subjective improvement for one signal, but the same filter may also result in a subjective worsening for another signal.


1) ... which is fine (though not necessarily so by the standards which apply in a market segment which has decided that expensive cables constitute a good solution and a decent EQ will never do ... but I digress), if they are by and large improving or not changing anything, and an algorithm may always provide for letting some signals through unaltered.  I assume there will be many signals where there is a theoretical case for improvement (though not necessarily any gain in practice, given the frequency range where it will kick in) - for example, if software can identify fairly reliably a string quartet, then one has a certain knowledge of where one should expect to find overtones (and thus, given the digitizing algorithm, their aliases).

2) And I wasn't restricting to that (basically because I have not read through all the marketing claims in detail); if back in the day the signal was digitized using a filter that started rolling off below 22.05 (e.g. http://xiph.org/~xiphmont/demo/neil-young.html#toc_o , but you know all that ...) then at least the frequency response part could easily be corrected (I didn't claim it would be audible ... and don't tell the audiophiles it is just an EQ).

Meridian Audio's new... sub-format called MQA.

Reply #173 –
Quote from: Porcus on
if back in the day the signal was digitized using a filter that started rolling off below 22.05 (e.g. http://xiph.org/~xiphmont/demo/neil-young.html#toc_o , but you know all that ...) then at least the frequency response part could easily be corrected (I didn't claim it would be audible ... and don't tell the audiophiles it is just an EQ).

Agreed.

Meridian Audio's new... sub-format called MQA.

Reply #174 –
Quote from: pelmazo on
there's an implicit and tacit admission in the MQA scheme: That 16-bit audio has enough dynamic range to sacrifice some of it for other purposes.


Do they really only deliver 16 bits including whatever extra information? Even if they use SPDIF, there is enough room for 24-bit audio in that 32-bit wordlength. And that is uncompressed - compressing they can effectively get more.