MQA at CES

[Miska]

I would like to hear the old version of the 2L files, and the new version. Maybe 2L could be convinced to make that available. Would seem one of the best chances to gauge what MQA can do as 2L had all the old hardware available to reverse engineer what it does to the signal. Of course you will need an MQA capable DAC to do the comparison.

 

Of course you need only decoder, just like HDCD decoder and then you can use any DAC for playing it back. I'm not going to trust it at all if I cannot run file encode - decode cycle and analyze the results against original.

 

MQA gives me personally a strong flashback of MLP from the DVD-A times...

 

Making a similar free and open source codec wouldn't be too hard. I was today thinking if Xiph would be interested on such.

[Miska]

Maybe MQA brings a little something to the table signal quality wise, it brings a very clever scheme of reducing file size without compromising sound quality which I see as its real benefit. As to it being an obvious or substantial increase in general sound quality I find that very unlikely to be the case.

 

I smell DRM and someone seeing lot of $$$ for IP licensing income (just like MLP)...

 

I don't understand why file size would matter at all, especially compared to 20-bit 96 kHz FLAC (same dynamic range as MQA).

 

It cannot really work losslessly for content that has busy top octave. It just relies that there's almost nothing there and then crams it in the 4 LSBs of 24-bit data while reducing dynamic range of the lower octave to 20 bits to make space for it.

 

My personal level of interest for MQA is at most 0. I see no benefits, only disadvantages.

[Miska]

While I don't have the knowledge to do such a thing it doesn't in principle seem beyond doing. I don't believe you could get a result equivalent to actually having used modern better equipment with good filtering, but an improvement seems conceivable. Yet again, it would be rather trivial if as good as advertised to do a before and after demo which everyone would hear and recognize the benefits. Why would you not do such an impressive demo rather than the apples and oranges demos they have done so far?

 

It's the usual apodizing filter stuff that Meridian has been doing for a while and also available in HQPlayer and some other places too. They just do it only up to 96 kHz while I go straight to the delta-sigma modulator output rate, minimum of 2.8 MHz and up to 24.576 MHz with current DACs on market.

 

Doing analysis of the ADC just confirms that it works as expected, but I've used the ADC and DAC chip datasheets as well as real world tests as basis for my filter design. I just don't care nor have huge marketing machinery to make such normal engineering stuff sound larger than life, my marketing budget is whopping 0€ (I don't go to shows like CES either, I rather use the money for R&D hardware instead).

[Miska]

Though I am not sure how you could get the access to what MQA does to actually make that comparison.

 

That's the primary problem. But at some point, there will be material with & without MQA and then one can compare and see.

 

I noticed that the 2L demo tracks seemingly don't have the corresponding non-MQA FLAC available so one cannot run diff. There's just 44.1/16 FLAC and 44.1/24 MQA FLAC. I'd much rather have 96/24 non-MQA FLAC or even 44.1/24 non-MQA thank you...

https://shop.klicktrack.com/2l/468051

[Miska]

But I have also heard new recordings they did themselves (controlling the entire A/D and D/A process and incorporating MQA) that sounded stunningly good. Is that because in those cases they can do a more precise matching of their apodizing filter through the ADC and DAC process?

 

I would say more likely to be due to good recording and mastering rather than any particular process. At least I would like to compare a true native non-MQA 192/24 content to MQA-content to hear and see how it changes. I would rather have those new recordings in 192/24, DXD or DSD128/DSD256 rather than having it transformed through low-rate PCM while trying to fix fundamental problems of low-rate PCM.

 

Since native DSD doesn't have digital filters involved, it fundamentally doesn't have related problems either. For lot of content content 192/24 is enough that there ends up being almost nothing to be filtered out and at DXD that is almost certainly the case.

 

Would it be something like setting up specific HQPlayer settings for each recording based not only on matching my DAC's chip but also what was in the original ADC?

 

Somewhat similar yes, mostly between choice of poly-sinc vs poly-sinc-hb/closed-form. But based on my experience the selection is also driven based on type of source material (classical vs pop/rock vs jazz/blues). IOW, if you start fixing the source (apodizing filters), you also want to select what aspect you emphasize in that process based on demands of the source content.

 

Piano material is bad for such testing, because piano doesn't really produce overtones reaching much beyond 20 kHz (when recorded from distance, rather than inside the piano). Unlike violin which produces plenty of overtones, but which is not producing transients, so it is not sensitive to transient time domain issues. For testing wide band overtones and transients together, percussions, glockenspiel and such are suitable.

[Miska]

I meant that looking at ultrsonics the level of the sound is so low that were we to encode only ultrasonics by splitting the band with a filter, you could often use the LSB(least significant bit) and maybe the next three bits. So even in a 16 bit signal the remaining 12 bits of more significance would be zero. MQA does something just like this and that lets them encode ultrasonics with much fewer bits.

 

This makes the assumption that the spectrum in ultrasonic band is like that. And usually, with average material when averaged over time it is. However with transients from percussions that is not the case for the transient period. The 50 kHz spectrum usage can be almost flat out for short period. So the average is not good for that case, only for content that doesn't have transients. So you may bring back transient distortion yet again by trying to save bandwidth (for no good reason).

 

I don't know precisely how FLAC handles this though I don't think they split the bandwidth that way to encode differently. Was does happen is you can see that FLAC with 44 or 48 khz sample rates typically compress to maybe half the size and perhaps not even quite half the size with some music. As you compress higher sample rates that have very few bits exercised in the ultrasonic range FLAC file sizes start to go down as a percentage of the original. Often at 96 khz rates the FLAC will be 45% or even a bit less file size. You can take those files and do your own 30 khz filter and the FLAC size drops a few percent further. So it would appear to me, that with such low levels of ultrasonic energy you end up with some additional redundancy which compresses smaller even when much of it is noise.

 

What I was thinking is to split the spectrum, cut the base band to 20-bit (with TPDF or noise shaping) and then take the top octave and encode it through Vorbis by using linearly descending psychoacoustic coefficients to constant bitrate of the remaining 4-bit bitrate and move that to the 4 LSBs of the 24-bit data. Then feed the entire "half-rate" data through FLAC. Or alternatively you can run more fancy adaptive algorithm to determine SNR of the source and choose 16 - 20 bit split for the base band vs 4 - 8 bit split for the top octave.

 

Now you have simple, free and open variant of MQA. (without ever reading a single word from any patent)

 

P.S. Extra bonus for using playback gain feature of Vorbis for temporal scaling of the amplitude range!

[Miska]

Yes, well, I do not think that the ultrasonics in a high rez recording, even with PCM, are in fact zeroes. They are, I believe, low level random noise. Hence, they will not compress as neatly and tidily in FLAC as you seem to think. That is true even if filtered, unless that filter is pretty drastic and potentially audible, even at 30k.

 

There is quite significant amount of relevant harmonic content up to 50 kHz in ordinarily recorded (with good microphones) content for strings and such. For percussion transients, the bandwidth can exceed 100 kHz without too much trouble, if the microphone is just up to the task.

 

That is my reading of the main thrust of MQA, particularly for its version of encoded, compressed hi rez, assuming an MQA DAC (or possibly pre-DAC MQA processing) is used.

 

I don't see any point in having "MQA DAC". DAC's purpose is to just convert sample to analog and that's all. For example the Mytek DAC currently having MQA support is using stock standard ESS Sabre DAC chip. Nothing special on that front. All the processing can be performed in software before the DAC if one wants to.

 

For apodizing filters one doesn't need MQA though.

[Miska]

If I understand correctly, if one does not have a MQA DAC, MQA files will play at only CD resolution.

 

What I don't understand is who would buy MQA files instead of normal hires FLAC or DSD download? Is download time or the space needed really an issue to someone?

 

If MQA has some magic processing, it could be all performed before encoding the thing as FLAC. Heck they could encode the "fancy" result as DXD FLAC. Then the listener doesn't need to have any special decoding stuff at their side.

[Miska]

Distribution is part of the goal and that means streaming without substantially increasing costs. Lossless streaming at high resolution sans Origami would not achieve that goal. Having the technology painlessly wrapped into your Tidal subscription does.

 

Netflix is streaming 4K video at reasonable monthly cost. And the bitrates we are talking about in that context are much higher than 192/24 FLAC compressed audio streaming would be (even if it were 5.1 channel).

 

So bandwidth usage in regards to audio is nonsense.

[Miska]

Very few people. The partnership with Tidal is more important and that's where bandwidth matters.

 

Why does it matter so much there? 5.1 channels at 192/24 FLAC would be still about half of what 4K Netflix or YouTube consumes...

 

Now the perfect next move from Apple would be to announce their music service at 96/24 ALAC... (I wouldn't be very surprised to see that in near future)

[Miska]

From what I gather streaming is not yet a highly profitable business model. I'm guessing bandwidth matters quite a bit to Jay-Z's accountants. And it certainly matters to consumers on restricted mobile accounts as well.

 

They can offer offline tracks like Spotify for those with restricted mobile accounts, that can be downloaded on unrestricted connection.

 

For the price of MQA-enabled hardware, I pay years of unrestricted LTE subscription (20 €/month for 50 Mbps here).

[Miska]

Please note that Netflix is providing VERY limited and compressed 4k!

 

Sure, it is 16 Mbps H.265 encoded video. But you get plenty of FLAC resolution at 16 Mbps. Uncompressed stereo DXD (352.8/24) is 16 Mbps.

[Miska]

Anyone state that you need a MQA DAC in order to decode MQA content, is not correct.

 

So they could decode it before streaming and delivery and deliver just ordinary standard FLAC, nothing of MQA needed at playback side.

[Miska]

Download here Miska and test it out

High Resolution Music DOWNLOAD services .:. FLAC in free TEST BENCH

 

Edit I was to quick. No MQA download yet. But at least DXD.

 

I know and have those files. So I still don't understand where you need MQA. They have the MQA files here:

https://shop.klicktrack.com/2l/468051?

 

Now tell me why there's no non-MQA hires (already decoded) download available there to compare? MQA version is remastered and upsampled version of the original:

Recorded in 44.1kHz/16bit by Lindberg Lyd, Norway, May 1993

[Miska]

You are a bit wrong, but also right. The standard SW for decoding MQA has a general DAC "signature". This I know for sure, but I do not know if the same signature will be used by everyboddy for MQA streamers and MQA SW players.

 

So normally if you like to go the very last mile, you need to be DAC spesific in decoding. That will normally be put into the DAC SW. However you could of cause make MAQ SW player's for the different DAC's available. I assume MQA will not deny you that.

 

MQA is only PCM output so it can be only half-way solution to anything. And if it runs on mobile phones and inexpensive DACs, the DSP process cannot be very fancy or high resolution.

 

 

But Miska, you that are a commersial participant, why not email MQA as they request on their website, and offer the HQPlayer with MQA functionallity or as an addon ?

 

In addition to nobody even asking for it, for three reasons:

1) They would want some serious $$$ for it

2) I'm not interested (I'm not interested on AAC or MP3 either, so it is not just MQA)

3) There's no way for me to make verification that it doesn't negatively impact sound quality/technical performance

 

People have been sometimes asking me for MP3 support in HQPlayer. But how much extra people would be ready to pay for the feature? As an example MP3 decoder patent license minimum yearly fee is $15000. That divided by yearly sales volumes of HQPlayer would mean significant increase in HQPlayer license price. Sure it is not much problem for Onkyo or Pioneer, who sell maybe hundred thousand or million devices per year, for them it is peanuts.

 

But I could add Vorbis support for example if really needed.

 

For me MAQ has only value when I use Tidal. Not sure if I can play Tidal through your player though ?

 

I'm listening Tidal through my player, although I don't use it much. I mostly listen Tidal in car and while otherwise on move. Through a mobile phone.

 

In some future I may purchase MAQ files as well

 

But why would you do that instead of FLAC, WAV or AIFF that you know you will be able to decode and losslessly convert to some new format even 20 years from now. Even if the companies and products go away.

 

To me, playback hardware and software is something that comes and goes, it is pretty small investment afterall. But music collection is expensive and hard to replace. That's why I have put so much effort into my player to make my RedBook rips sound as good as possible. Those won't be replaced with something else. I have the material in FLACs now, and the silver discs in the storage. And I know that in 20 years from now, I can convert that content to a new lossless container and still keep listening.

 

If you have MQA download and MQA begins to disappear from the market/products, how do you convert the contents to something else, like FLAC without losing the "hires" resolution you paid for?

 

Thanks to some clever reverse engineering I've been able to rip my few HDCD discs as 44.1/24 FLACs (note: you need to disable the gain option in dbPoweramp plugin, otherwise it will decode it incorrectly). But that capability wasn't given at the time of purchase. Luckily I didn't pay anything extra for the HDCD.

[Miska]

I have made a small misinform. The compressed file will keep it's original bit. So in most 2L cases that means 24 bit not 16 as indicated above.

 

But in case of the currently available demo track it is 44.1/16 which is the original recording resolution?

[Miska]

Well if decompressing in streamer or SW player gives 24/352,8 it may be OK, but does it ?

My DAC has max 24/192, so I would not want 352,8.

 

I always try to buy the original resolution, what ever it is. Regardless of my DAC(s) capabilities. And let playback software convert the content to a format most suitable for the DAC I am currently using.

 

There's no way I'm going to purchase the same content again at different resolutions every time I change DAC. I already have at least 10 different DACs with different capabilities. But I will only buy one format of each album.

[Miska]

Where is the logic behind a 24/44.1 FLAC file coded with MQA is decoded as 16 bit FLAC.

 

I did some analysis and the result is quite a bit less than 16-bit worth. While the source FLAC file is pretty big. Here's some of my initial results:

 

Some analysis and comparison of MQA encoded FLAC vs normal optimized hires FLAC - Blogs - Computer Audiophile

 

In short, my result was that if you optimize the encoding to a standard FLAC (without MQA), the file is smaller than the MQA encoded one...

[Miska]

As a partner they were likely provided an SDK with functions they could incorporate into their code. They obviously went ahead and did that. The issue here is whether that implementation will earn them the right to call their product MQA certified.

 

You can find some answers here, on their FAQ:

MQA | For playback providers

[Miska]

+ Just to make clear. In the piano and the magnificat comparisons, I also had the DXD tracks gone through the same resample logarithm. So at least the difference had nothing to do with the resample log.

+ I have removed the DXD tracks and put the MQA in their place.

 

Do you have MQA decoder in your system? If not, the resolution you get from MQA file is quite a bit less than RedBook. While DXD has way more resolution.

 

I will quit my Tidal subscription if they start using MQA for all their "HIFI" streams, because it is much worse than RedBook. Then I can as well keep on using Spotify (which I'm also subscribing) at about same quality, but half the price and more content.

[Miska]

"Much worse than Redbook"? You're really losing credibility here, at least with me. Obviously a well mastered Redbook file is not "worse" than a DSD256 file from a poor master and that's essentially the argument you're making here. Makes no sense.

 

Did you check the blog post I made? I compared the MQA FLAC to plain normal 44.1/16 FLAC and it is much worse. Less than 16 bits left.

 

Se it yourself...

 

16 megabyte MQA FLAC (from 2L):

 

vs

 

6.2 megabyte RedBook FLAC (made with my own tools from the original 2L DXD):

 

vs

 

13 megabyte 120/18 FLAC, optimized to not loose anything from the source (made with my own tools too, from the original 2L DXD):

 

 

Where's the space saving vs quality!?

[Miska]

I read that and didn't understand how you determined it was less than 16 bit. Could you please explain?

 

From the amount of background noise. If you compare the RedBook and MQA version, the MQA version has background noise at best at 16-bit level at best on some frequency areas, but above 15 kHz it begins to drop significantly below that of 16-bit. So compare noise levels especially between 15 - 20 kHz.

[Miska]

Does the above tell us anything about how a properly MQA decoded file will sound??

 

No, I only care how it performs without MQA decoding, as alternative to RedBook FLAC. But at least so far, I can create a standard (no secret sauce) FLAC that has no quality degradation compared to the original DXD and it is still smaller than the MQA file!

 

I don't want to buy MQA equipment just in order not to get quality degradation on Tidal!

[Miska]

They also say they can undo damage of time smearing filters and reproduce what the signal was before it was digitized then digitize it anew with filters that smear time less.

 

That's called apodizing filter and it's nothing new. They have been themselves doing it for a long time in their equipment.

 

What you cannot do is to put back something there that the filter or quantization has removed, no matter how much you inspect the original equipment. If the information is not there, it is lost for good. For hires, you need hires.

[Miska]

OR could be an artifact of these particular recordings in this A/B comparison, so more of these will obviously need to be done...

 

I downloaded and checked every single of the MQA files and the corresponding DXD file from here:

2L High Resolution Music .:. free TEST BENCH

 

And every single of those exhibit the same pattern...

 

 

I do some guessing, but I doubt the MQA decoder expands the dynamics of the baseband down, even less removes the HF-noise. It probably just expands the frequency band up leaving the base band intact.

 

After some speculation and playing with two gentle noise shapers I have, I can cut the original to 352.8/8 with one and 352.8/10 with the other before the 0 - 22.05 kHz band begins to look like the MQA FLAC. Meaning that I would be left 14 - 16 bits per each 24-bit sample for encoding the band above 22.05 kHz...