Meridian's mysterious MQA site.

[esldude]

Okay for those wondering how this could work, you can try the following experiment. Take any digital audio file you have in wav and convert it to FLAC. Usually you will get a file about half size. Now use some software to reduce the level of that same file by 50 db. Save that FLAC. You will get a far smaller file. Perhaps as much as an eighth as large.

 

Now in this case you lost some resolution by lowering the level in the first place. However imagine if you could do one thing. Split a file with 96 khz bandwidth into below 20khz and 20khz -96khz. With most music the upper part of that split will be very low in level. Now if you used FLAC for two halves of that file the upper frequencies will get compressed losslessly to a quite small size. If you could put those back together upon playback you would lose nothing. Yet the total bits would be considerably less than if you simply took that 96khz bandwidth file and saved the whole thing straight to FLAC.

 

Another way to see it is perhaps easier. If your level in a given frequency band has dropped to -76 db and lower, if it were a 16 bit signal you could encode it fully using only 3 bits and noise shaping. Many recordings do this in the upper frequencies. So using all 16 bits for this is a waste. A much smaller number of bits will fully encode what is there in upper frequencies.

 

 

This isn't the whole thing with MQA, but it is one part of it.

 

Another part is I believe they are changing the file to a SUM and Difference file. Usually those are going to compress smaller than the stereo files will. So that is another part.

 

On top of that they aren't even keeping all the bits for upper frequencies, but only the difference in signal levels between 20khz and above 20khz. That also saves bits. So there are several things going on to make this work.

 

Now you have watch what Bob Stuart says. There is Robert Stuart the researcher at the AES, and Bob Stuart one of the principles of Meridian Audio a company. I am not bashing Bob too much, but Bob sometimes translates things into marketing speak. Somewhat necessary for any business to do. When he says they store timing information he isn't really referring to some bits allocated to some other form of encoding timing. All he really means is they keep enough bandwidth so they can use apodizing filters with gentler slopes which are more precise in the time domain than steeper filters would be.

 

Further I also believe it isn't quite fully lossless. I believe they are actually only keeping enough bandwidth to work to somewhere around the 30 khz range. You could say it is lossless to anything that is there to 30 khz. Above that it is somewhat lossy with some info that may with some noise shaping DSP let the filtering work as if it were quite gentle indeed. He has referred to it as Perceptually Lossless. This is also some of what he means about timing info. Not sure if you would classify that as lossy timing info. Better than not having it, but not the complete original.

 

So it is being pitched as effectively lossless, as allowing 384 khz sample rates at 1 mbps. Yet it is sort of that, but not really. It appears to be Mr. Stuart's opinion that timing info in higher sample rates which allow better gentler filtering are the benefit. That response past 20 khz isn't heard except as a side effect of filter effects. So he is giving you the flat 20khz and the filtering performance as if it really were a high sample rate signal. Doing so in a fairly compressed data stream.

[YashN]

Interesting info, elsdude.

[spdif-usb]

Okay for those wondering how this could work, you can try the following experiment. Take any digital audio file you have in wav and convert it to FLAC. Usually you will get a file about half size. Now use some software to reduce the level of that same file by 50 db. Save that FLAC. You will get a far smaller file. Perhaps as much as an eighth as large.

 

Now in this case you lost some resolution by lowering the level in the first place. However imagine if you could do one thing. Split a file with 96 khz bandwidth into below 20khz and 20khz -96khz. With most music the upper part of that split will be very low in level. Now if you used FLAC for two halves of that file the upper frequencies will get compressed losslessly to a quite small size. If you could put those back together upon playback you would lose nothing. Yet the total bits would be considerably less than if you simply took that 96khz bandwidth file and saved the whole thing straight to FLAC.

 

Another way to see it is perhaps easier. If your level in a given frequency band has dropped to -76 db and lower, if it were a 16 bit signal you could encode it fully using only 3 bits and noise shaping. Many recordings do this in the upper frequencies. So using all 16 bits for this is a waste. A much smaller number of bits will fully encode what is there in upper frequencies.

 

 

This isn't the whole thing with MQA, but it is one part of it.

 

Another part is I believe they are changing the file to a SUM and Difference file. Usually those are going to compress smaller than the stereo files will. So that is another part.

 

On top of that they aren't even keeping all the bits for upper frequencies, but only the difference in signal levels between 20khz and above 20khz. That also saves bits. So there are several things going on to make this work.

 

Now you have watch what Bob Stuart says. There is Robert Stuart the researcher at the AES, and Bob Stuart one of the principles of Meridian Audio a company. I am not bashing Bob too much, but Bob sometimes translates things into marketing speak. Somewhat necessary for any business to do. When he says they store timing information he isn't really referring to some bits allocated to some other form of encoding timing. All he really means is they keep enough bandwidth so they can use apodizing filters with gentler slopes which are more precise in the time domain than steeper filters would be.

 

Further I also believe it isn't quite fully lossless. I believe they are actually only keeping enough bandwidth to work to somewhere around the 30 khz range. You could say it is lossless to anything that is there to 30 khz. Above that it is somewhat lossy with some info that may with some noise shaping DSP let the filtering work as if it were quite gentle indeed. He has referred to it as Perceptually Lossless. This is also some of what he means about timing info. Not sure if you would classify that as lossy timing info. Better than not having it, but not the complete original.

 

So it is being pitched as effectively lossless, as allowing 384 khz sample rates at 1 mbps. Yet it is sort of that, but not really. It appears to be Mr. Stuart's opinion that timing info in higher sample rates which allow better gentler filtering are the benefit. That response past 20 khz isn't heard except as a side effect of filter effects. So he is giving you the flat 20khz and the filtering performance as if it really were a high sample rate signal. Doing so in a fairly compressed data stream.

The only marketing strategy that Bob Stuart is using is that of the old mousetrap theory, i.e. if you put strong enough cheese in, eventually someone's gonna bite.

 

AES Los Angeles 2014 » Paper Session P14: Perception: Part 2 (Scroll down to where it says "P14-7 A Hierarchical Approach to Archiving and Distribution—J. Robert Stuart, Meridian Audio Ltd. - Huntingdon, UK; Peter Craven, Algol Applications Ltd. - London, UK".)

[jhwalker]

The only marketing strategy that Bob Stuart is using is that of the old mousetrap theory, i.e. if you put strong enough cheese in, eventually someone's gonna bite.

 

AES Los Angeles 2014 » Paper Session P14: Perception: Part 2 (Scroll down to where it says "P14-7 A Hierarchical Approach to Archiving and Distribution—J. Robert Stuart, Meridian Audio Ltd. - Huntingdon, UK; Peter Craven, Algol Applications Ltd. - London, UK".)

 

Yes, that's the method we've been discussing - what of it?

[spdif-usb]

Yes, that's the method we've been discussing - what of it?

Just that there's no "translating things into marketing speak" involved in any way at all.

[audiventory]

Nice test... but again not for human ears, for measuring programs instead. So, we humans can ignore the conclusions.

Maldur,

As show my (and not only my) experience usually improving of measured values directly correlate with ear's perceptions. Need remember what is not enought concider alone measuremens. Need concider it together only due improving one feature can lead to deterioration of other.

[YashN]

I see no particularly weird marketing strategy used here. We know about the existing of the paper and the presentation at AES.

 

The only marketing strategy that Bob Stuart is using is that of the old mousetrap theory, i.e. if you put strong enough cheese in, eventually someone's gonna bite.

 

AES Los Angeles 2014 » Paper Session P14: Perception: Part 2 (Scroll down to where it says "P14-7 A Hierarchical Approach to Archiving and Distribution—J. Robert Stuart, Meridian Audio Ltd. - Huntingdon, UK; Peter Craven, Algol Applications Ltd. - London, UK".)

[monteverdi]

John Aktinson lost his cool http://www.stereophile.com/content/ive-heard-future-streaming-meridians-mqa

[esldude]

John Aktinson lost his cool I've Heard the Future of Streaming: Meridian's MQA | Stereophile.com

 

Had not seen the Shannon diagrams listed there. It is interesting to note, that all but the very tiniest amount of info above noise is captured by 96/16. That turned into FLAC format is 1.5 mbps. If we have the tiny little bit not captured at 96 khz is it really audible? If so, it seems it would be a tiny, tiny difference. Still if we can get the extra bit with two thirds the data rate, I have no complaints.

 

Further if it did become the defacto format, having the one format everywhere for everyone on everything with no loss of fidelity, then color me very pleased. I would like for them to have heard a 44 or 48 khz sample for comparison though.

[Don Hills]

John Aktinson lost his cool I've Heard the Future of Streaming: Meridian's MQA | Stereophile.com

 

The money shot:

... MQA intends to guarantee the provenance for the high-resolution masters—no more upsampled CD masters masquerading as "hi-rez." ...

 

I wonder how they are going to guarantee that? License the encoder on condition it's only used for true high-resolution sources?

 

Spot the problem:

... a track by Metallica ... With MQA, I heard far more definition and separation between instruments and lines in the lighter beginning of the track. When the band went all out, I was astounded at how much more colorful and undeniably brutal the sound was with MQA engaged. The full impact of heavy metal came through in ways that I would have thought possible only in live performance. And this, mind you, was at a volume level that, for the sake of my ears, had been turned down several notches from my initial listen without MQA. That the music could be lower in volume, yet sound meaner and more vile in the best of ways, confirmed the musical dividends that MQA can deliver.—Jason Victor Serinus

[esldude]

Maybe he heard the Guitar Hero version of Death Magnetic. It is far, far better than the version on the CD.

 

[Boris75]

MQA sounds like mp3 done right: keep musical information, discard noise.

[YashN]

It isn't a lossy compression like MP3.

[ted_b]

The money shot:

 

 

I wonder how they are going to guarantee that? License the encoder on condition it's only used for true high-resolution sources?

 

Spot the problem:

 

"MQA engaged"? Hmm...so let's hope spdif-USB is right and it is additive (i.e the MQA disengaged signal is just the regular track).

 

I am not against discovery and new things, and Bob and team had done a great job with the technical side of DVD-Audio (MLP); the marketing failed, but that's hardly the tech fail. But if this new encoding requires decoders and new hardware the industry may have a tough time selling that to a public who has shown an allergy to new "standards".

[jhwalker]

"MQA engaged"? Hmm...so let's hope spdif-USB is right and it is additive (i.e the MQA disengaged signal is just the regular track).

 

I am not against discovery and new things, and Bob and team had done a great job with the technical side of DVD-Audio (MLP); the marketing failed, but that's hardly the tech fail. But if this new encoding requires decoders and new hardware the industry may have a tough time selling that to a public who has shown an allergy to new "standards".

 

Per the article, one of the benefits of the product is that it is backwards compatible with devices that do not include MQA decoding - all the extra encoding is tucked into areas of the file that are not normally read (i.e., past the 16-bit threshold), so the files are played as "normal" 16/44 files on those devices. Only the devices that are MQA-enabled will be able to see / read (and, of course, decode) the extended data.

[ted_b]

Yeah, after posting I got more educated. Sorry...my bad.

[Boris75]

It isn't a lossy compression like MP3.

 

It is.

[esldude]

It is not lossless. But it isn't lossy in the conventional sense. Some comments from Meridian and reading the patent indicate it is a hybrid of sorts. For some portion of the bandwidth (exactly how much bandwidth isn't clear yet) it is fully and losslessly encoded. Some other portion of the upper frequencies is a lossy encoding. The lossy part is being done in a way not possible for us to hear, but well enough to benefit the final sound (which sounds contradictory of course).

[Boris75]

Not at all.

 

Now, if you can't read simple English explanations, I think Christmas comes early for you and you win a place on the Ignore list.

 

Your sarcasm is very funny, given that you are wrong. Please re-read my posts slowly, and then elsdude's post, and perhaps you'll finally understand.

[beanbag]

So it's like HDCD, except in frequency instead of amplitude?

[DanSmedra]

This HiFi+ article goes a bit deeper. In concept, it's intriguing.

 

MQA - It's about time! Meridian's new MQA codec brings neuroscience to music.

[esldude]

Description of the process from the patent application here:

GB2013051548 DOUBLY COMPATIBLE LOSSLESS AUDIO BANDWIDTH EXTENSION

 

Block diagram from the patent.

 

[Don Hills]

Maybe he heard the Guitar Hero version of Death Magnetic. It is far, far better than the version on the CD.

 

 

Maybe, but irrelevant. Did you see the problem?

[Boris75]

In his CES report, our host confirms that, when filtering out marketing claims of MQA being "audibly" lossless, MQA definitely seems to be lossy.

 

Another reason I remain unconvinced is that, I believe, MQA is NOT lossless. Meridian claims MQA is lossless, but I believe it only claims audibly lossless, not lossless in the normal use of the term. I checked with a couple engineers at the show and each confirmed MQA wasn’t lossless in the traditional sense of the word.

[shadow ewe]

I wouldnt lose any sleep over this "MQA". The fact that it is going to charge the industry for using makes it seem fishy.

 

I still remember this:

 

Meridian Lossless Packing - Wikipedia, the free encyclopedia

 

So I am going to steer clear of this hype.

 

It actually makes me quite sad that TAS and Stereophile did not reference the past to shed more light and instead go "ga ga" over MQA and tout itt as revolutionising PCM ad infinitum.

 

Give it 12-18 months and no-one will be talking about MQA. That is my prediction.