Audio Asylum Thread Printer
Get a view of an entire thread on one page
|For Sale Ads|
In a recent post here [on Digital Drive] by Charles Hansen he says:
"MQA apparently asserts that the at least 6.8 (and possibly as much as 11.8) of the lower-order bits are inaudible."
This is in reference to the lower order bits being re-purposed by the proprietary MQA encoding scheme ... per Hansen:
"However the seven or eight Least Significant Bits (LSBs) of the container are used to store the "folded" dual-rate audio data (encoded losslessly) and the quad-rate audio data (encoded with lossy compression). They *replace* the low-level bits in the original 24-bit file. . reducing the resolution of MQA below 24 bits."
I will not concentrate on the comment "... reducing the resolution below 24 bits", or even that MQA resolution is "... limited to a maximum of 17.2 bits" (same post). Those who would like to contend those assertions should have at it directly with Hansen ... and in advance, good luck to you, you'll need it!
Rather Hansen's comment triggered another thought ... *if* the lower order bits are relatively insignificantly (1) why not just truncate [the lower order bits] and stream the remainder? ... with the missing bits filled in on the receiving side (e.g. some dithering scheme or whatever).
So we still lose the low order bits ... no change fundamentally since that are effectively lost already with MQA ... but forego the need for the fancy "origami" folding, including the lossy compression of 4FS audio data.
(1) To be clear Hansen make no claim the lower order bits are insignificantly, in fact he casts doubt on just that. However if the idea of tossing the LSBs is a non-starter then MQA is a non-starter ...for very same reason!
Yet another format variant.
I'll revisit this in 10 years... IF it's still around
I've read this entire thread up to this moment, and haven't seen any references to Meridian's official technical claims/processes.
Surely, they must have a "white paper" or even a technical paper that someone here has read or seen. Can anyone point me to such a publication?
Benchmark is a good start, their statement is informative and they provide useful links at the end, including to MQA material, e.g. patent.
While it is true that there is no such thing as a free lunch, the converse also holds true. While MQA discards the lower bits in the baseband of a true high-res (eg, 192/24 LPCM) source file, they are replaced with losslessly compressed bits from the dual-rate band and lossy compressed bits from the quad-rate band. Both of these "foldings" represent data that is correlated with the music. Therefore even though an MQA file is a lossy representation of the original high-res file (as information theory dictates), it does contain more information than would be by simply truncating the bits of a single-rate file. In other words, a 192/24 file fully hardware decoded by MQA will have a maximum resolution of ~192/17.
If one believes that bit depth is the most important parameter, this would not be much of an improvement over a standard Redbook (44/16) file. On the other hand if one believes that the sampling rate is more important (as do I, as it allows for much gentler, less destructive filters), then there is a benefit. However why not have your cake and eat it too? At the Munich show Qobuz announced a third tier (above Redbook FLAC Sublime) called Sublime+ that allows for full high-res streaming of files up to 192/24. An interesting note is that this costs more money. The implications is that those who set the pricing (largely the record labels and copyright holders) apparently value true high-res files more than MQA files, as the price for MQA files is the same as the price for Redbook files. Possibly a case of getting what you pay for.
MQA's argument is that the bits that are discarded (and replaced by other correlated bits) in order to reduce the file size are inaudible. That is precisely the same claim that was made regarding MP3 - sonically indistinguishable. My most recent technical analysis (with minor corrections from previous posts being noted in the paragraph beginning with "NB") is in this post:
As always my posts reflect strictly my own opinion and not that of my employer, family, or book-club members.
Thanks for chiming into this thread. I read all your posts. I think you've probably read my editorials, but if not, over a year ago I began asking many "questions" about MQA.
Some of those questions revolve around the supposed "correction" being done for the source A/D. That isn't being talked about here. Do you have comments on that? Obviously, questions abound about if they're actually capable to do that -- and what if there was more than one A/D used (including cascading ones) -- but I'd like to see your comments on that part of the process.
To be perfectly honest, I unfortunately missed your first editorial regarding MQA, from April 2016:
I did see your more recent editorial comment (linked below). IMO both were *excellent* explorations of the topic, and I highly recommend that all AA readers take a look.
To answer your specific question, I believe that I covered that in my reply to Tunenut in this post:
Please let me know if I missed something or if you would like additional clarification. Thanks!
As always my posts here are strictly my own opinions and do not necessarily reflect those of my employer or ex-mother-in-law.
write all your own sig lines fresh each time, or do you have some sort of auto-generator? They're very funny. Thanks.
> > do you have some sort of auto-generator? < <
That's a good idea for an app! Sort of like the old "Magic 8-Ball".
Thanks for the kind words. I just write them off-the-cuff, as some (perhaps naturally) assume that my postings are a reflection of my employer's official company position, which is not necessarily true. The various non-sequiturs are my way of gently reminding readers not to jump to an unwarranted conclusion, and perhaps get a chuckle at the same time.
I realized afterwards that you had posted something about this -- I was looking at the wrong part of the thread.
In any event, this so-called correction, *to me*, is the most intriguing part of MQA -- if, in fact, it can be done. (The compression gets more irrelevant with the passing of each day.)
Regarding this correction, I'd think it would be VERY easy for the MQA folks to demonstrate by yanking out an A/D that "smears" what they correct and record their own demo tracks to show it in action. But the lack of comparison material is the most troubling aspect of this technology -- there's very little. That, to me, makes the whole thing suspicious.
> > Regarding this correction, I'd think it would be VERY easy for the MQA folks to demonstrate by yanking out an A/D that "smears" what they correct and record their own demo tracks to show it in action. < <
One would think so, yes? And if MQA could show this, it would seem that the record labels and mastering studios would be lining up to buy this special technology that can miraculously improve the sound quality of existing digital files. Just as everyone would line up to buy a cheap additive to your automobile fuel that would yield 200 MPG.
As always, these postings strictly reflect my own opinions and not necessarily those of my employer, big sister, or Santa Claus.
Yeah, this is where it loses me. If you're going to make bold technical claims about something in marketing literature, then that's one thing. But to make it to designers or people who know better then not be willing to show one bit of evidence to prove what you're saying -- even though it would be very easy to -- then that's another. If they did prove that, I'd be much happier. Till then...
> > But to make it to designers or people who know better then not be willing to show one bit of evidence to prove what you're saying -- even though it would be very easy to -- then that's another. If they did prove that, I'd be much happier. Till then... < <
It would seem that they are relying on listening tests to "prove what they're saying". The difficulty is it appears that none of these are true apples-to-apples comparisons. Specifically the MQA files are always played back through the MQA slow-rolloff minimum phase digital filter, while all other files are played through a different digital filter - almost always the stock one found in the DAC chip.
There are quite a few manufacturers who have taken the trouble to learn how to design and implement custom digital filters. The reason to spend the required time, resources, and increased parts cost is because those manufacturers believe they can build a better-sounding digital filter than the stock ones found in DAC chips.
The main conclusion one could reach of those that prefer the sound of MQA files would seem to be that MQA has created a digital filter that sounds better than the stock digital filters found in DAC chips. Which simply supports the notion that manufacturers that use custom digital filters also have the opportunity to create better sounding products.
I suppose that it would also be possible to implement a custom digital filter that sounds *worse* than found in a typical DAC chip. This would also imply that various custom digital filter implementations have varying degrees of sonic performance. Since the MQA digital filter is a custom (ie, not off-the-shelf) digital filter, it would be logical to conclude that some custom digital filters sound worse than MQA's solution and that some custom digital filters sound better than MQA's solution.
If one accepts this premise, then it becomes clear that it is possible (and possibly likely) that there are DACs that will play back true high-res files and sound better than MQA files, as the original high-res file will *always* contain more musical information (better source material) which is then fed through a superior-sounding custom digital filter. An obvious bonus of this scheme is that if one purchases a DAC with a superior-sounding custom digital filter that it will sound better with *all* digital files and sources, and not just the small number of titles that have been encoded in MQA (which activates the MQA custom digital filter).
As always, my postings here reflect only my opinion and not necessarily that of my employer or the currently hottest pop star.
I agree that the nonexistent apples-to-apples comparisons is a problem.
But this supposed "correction" they're talking about implies more than only the filter at the end. Supposedly, embedded with that MQA file, is information about the encoding A/D and that would then alter the behavior at the playback end to compensate for whatever errors were introduced at the start.
*IF* that works, I think it would be quite easy to demonstrate if they really wanted to.
> > But this supposed "correction" they're talking about implies more than only the filter at the end. Supposedly, embedded with that MQA file, is information about the encoding A/D and that would then alter the behavior at the playback end to compensate for whatever errors were introduced at the start. < <
I don't think that MQA is claiming that the encoding process also sends information that alters the behavior of the D/A converter. My understanding is that the MQA encoding process "corrects" ("de-blurs") errors in the existing digital file created by an imperfect A/D converter. Then on top of that, when a D/A converter manufacturer applies for an MQA decoding license, MQA will analyze the digital filter already used and adjust the coefficients in the added MQA digital filter to "compensate" for the "imperfections" in that already present digital filter. This filter is already present because the vast majority of D/A converters are built with chips that incorporate the digital filter function and the DAC function itself into one chip (to reduce costs).
On the other hand there are a few D/A converter manufacturers who separate those two functions into two separate circuits because one can build a higher-performance digital filter than the cost-constrained designs that are built into DAC chips. The manufacturers that have taken this path have invested significant resources into developing what they feel is the highest possible performance, and may not be overly excited about the prospects of a third party modifying the digital filter they have developed. And conversely, manufacturers that lack the resources to design custom digital filters may be very excited for a third party to supply them with a digital filter that may out-perform those built into DAC chips.
Pacific Microsonics' HDCD process has many similarities to the overall scheme of MQA, as both claim to solve "problems" existing in (then) current digital technology, and both also sell (sold) digital filter chips (or algorithms) they claimed out-performed those commercially available. With HDCD, PM claimed their process increased the limited resolution of the 16-bit CD to 20 bits. A close examination reveals that there was actually only about a 1-bit increase in resolution due to their compansion (compression/expansion) process. The other 3 bits of claimed resolution improvement were actually achieved via dither - something that is not patented nor proprietary and hardly exclusive to HDCD. The slight increase in resolution became moot when *true* high-resolution formats were developed, including both DVD-Audio and SACD.
If MQA claims to improve existing digital files by "de-blurring" (removing time-smear introduced by the digital anti-aliasing filters in the A/D converters used to create the recordings), there is nothing to stop others from also developing digital filters that accomplish the same goal. In fact many already have. Furthermore it is also possible to develop A/D converters that have no "blurring" that adds artifacts to the original analog signal, and hence would derive no benefit from "de-blurring". And in fact many already have.
The other benefit of MQA is that it reduces file size through a combination of lossy compression and discarding the lower bits in the original high-res file. The resulting loss of resolution is claimed to be inaudible, just as was the case with MP3. In both cases the falling cost of digital storage and the rising speeds of network connections has largely negated the need for such compromises.
> > *IF* that works, I think it would be quite easy to demonstrate if they really wanted to. < <
It would seem there are many ways MQA could easily dispel the confusion and controversy surrounding the process. Many (besides yourself) wonder why they have not.
As always, these posts are my personal opinions and not necessarily those of my employer or my golf caddy.
This comment from you is interesting: "My understanding is that the MQA encoding process "corrects" ('de-blurs') errors in the existing digital file created by an imperfect A/D converter."
As a so-called "end-to-end" process, my understanding that the "correction" wasn't done in the actual encoded file but, rather, at the end, in the DAC itself, along with adjusting for imperfections in the DAC.
With that in mind, if, in fact, they are correcting the original file and that MQA file has been modified for A/D imperfections then the the entire thing is even more lossy than I first thought. First off, as you mention, the lower bits are gone and some compression goes in to getting it all into the 24/48 constraint they are using. Then, if a process that corrects for timing of the A/D was applied, well, that plain and simply alters the original file. Just one of those MANY questions I had (and have).
Someone just brought a post to my attention (linked below). This is from Gordon Rankin, who designed the award-winning AudioQuest DragonFly series of USB DACs. The latest two models have been designed with MQA in mind and last week a firmware update was released that allows for some sort of MQA compatibility.
I must say that I don't yet understand this process. Remember that software decoding of the first "fold" was a surprise only announced when Tidal began streaming MQA in late January. This is the first I've heard of "custom filters for each song" and leaves me with more questions than answers:
1) If there is specific filtering done for each song, why not do it during the encode process?
2) The scheme is very reminiscent of similar schemes patented for the A/D process by Pacific Microsonics for HDCD and for the D/A process by Ed Meitner and assigned to Museatex for the BiDAT DAC. Both processes had two separate filters (anti-aliasing for the A/D side and reconstruction for the D/A side). The units would automatically switch coefficients depending on the signal level in the top octave. Is there more to this in MQA's implementation? How many sets of filter coefficients are available?
3) The DragonFly DACs will not normally decode signals higher than a 96kHz sample rate. My understanding is that the software decoding performed in the Tidal software only outputs a maximum sample rate of 96kHz. Does the DragonFly do anything more than fine-tune the filter coefficients?
It would seem that MQA is still under development. Interesting times. As always these posts are my opinions only and not necessarily those of my employer or previous employer.
> > if a process that corrects for timing of the A/D was applied, well, that plain and simply alters the original file < <
Yes, the original high-res file is altered in at least three ways:
1) The quad-rate audio data is compressed using lossy techniques.
2) The lowest 6 to 8 bits in all frequency bands are discarded to allow the dual- and quad-rate information to be "folded" underneath the audio data in the baseband. This is claimed to be "inaudible" as it is below the noise floor of the electronics. However it is well known that the ear/brain can distinguish correlated music 10dB or even 20dB below the noise floor of an LP, for example.
3) The so-called "blurring" in the original file is created by the digital anti-aliasing filter used in the A/D converter. This "ringing" is at a specific frequency - the corner frequency of the anti-aliasing filter - typically Fs/2. One concern about this filtering process is to ensure that new artifacts are not introduced.
In this sense MQA is like MP3 - both are lossy processes and the full original data can never be recovered.
All postings are strictly my own opinion and not necessarily those of my employer or my kids.
..and Doug, did you now that around the end of last year/beginning of this year, Meridian/MQA fired their US based PR firm? They now handle all inquiries direct. Why?,,, the audiophile press (you excluded!!!) has been their de facto PR service, at no cost.
> Doug, did you now that around the end of last year/beginning of this year,
> Meridian/MQA fired their US based PR firm?
Meridian's PR is indeed now handled from the UK. MQA's US PR is still
handled by Sue Toscano, who has been responsible for the account since
the start, first with Nicoll PR and now with her own company.
"But this supposed "correction" they're talking about implies more than only the filter at the end. Supposedly, embedded with that MQA file, is information about the encoding A/D and that would then alter the behavior at the playback end to compensate for whatever errors were introduced at the start.
*IF* that works, I think it would be quite easy to demonstrate if they really wanted to."
They can't and won't. Because in a word, it is utter bullshit.
And if we stretch our imaginations to their absolute limit, and suppose they they have unlocked the secret to flawless digital encoding, why not design an ADC that fixes ALL issues, this supposed "smearing" at the start.
The answer is clear.
...it's time someone added light instead of heat on this topic.
Thanks for your clear explanation. I understand that the folding of high bit-rate files into lower bit-rate files is but one aspect of MQA- and it is not the one that interests me that much, because, at best, you preserve the quality of the high bit rate.
There is also, as I understand, a claim of improving the quality through the use of processing the audio file to correct for time domain problems in the original recording. This seems as though it could be useful without any sort of new DAC purchase. Do you have any information or thoughts about that aspect of MQA? Thank you.
> > improving the quality through the use of processing the audio file to correct for time domain problems in the original recording < <
To the best of my knowledge the only way to do this is to insert a digital filter into the chain, typically an "apodizing" filter that has a lower cutoff-frequency than the anti-aliasing filter used in the original A/D converter. This will filter out any "ringing" on the recording - but the problem is that the new digital filter will impose its own ringing. The idea of the apodizing filter is that by using a minimum-phase filter, all of the pre- and post-ringing imparted by the A/D converter will be replaced with only post-ringing - thought to be less intrusive sonically.
There are at least two issues with this approach:
1) There are only a minority of recordings where this can be applied. Most early CD releases were transferred from analog tapes using a single stereo A/D converter. While it is conceivable that a digital filter could improve the sound of these recordings, that digital filter can be anywhere in the playback chain. Why insert a special one upstream before the point the file is distributed? It makes more sense to put a filter of this type into the playback DAC so that *all* recordings will benefit rather than a few that have been given a "special proprietary process". Then the end user would not be forced to purchase new hardware to play back these "new" specially encoded files.
2) For the last two decades (at least), most recording start in the digital domain. A modern recording may have several dozens of different A/Ds used to create the final product, each operating at different sample rates, often in from different manufacturers, and even in different studios. There is absolutely no way to "compensate" for all of these different A/D converter "signatures" at once with a single filter.
All postings are my own personal opinions and do not necessarily reflect those of my employer or the local dog-catcher.
The points you made above I raised months ago. And you are spot on.
"There is absolutely no way to "compensate" for all of these different A/D converter "signatures" at once with a single filter."
But there are some who would convince you otherwise.
Charles, why don't you send your DACs to Meridian so they can take it apart and get the blueprint, so you can have a nice shiny "MQA Ready" stick on every box that leaves the factory, once you pay the fee of course, and update the firmware? Lossly, not lossy...who cares.. don't you want to cash in on the lasted three letter acronym?
Sorry, if you don't jump on board the MQA train, your DAC models will no longer be "competitive"....
> > why don't you send your DACs to Meridian so they can take it apart and get the blueprint < <
The very first audio company to promote slow-rolloff "gentle" digital filters (to the best of my knowledge) was Wadia in the late 1980s. Wadia was formed by a group of digital engineers who left 3M to make their fortune in high-end audio. At that time there were only two companies in the world to have their own custom digital filters (made using DSP chips), Wadia and Theta. Everybody else used a digital filter chip designed by the semiconductor manufacturer, initially either Philips or Sony, later from Pacific Microsonics, later mostly from Burr-Brown or Analog Devices, and currently mostly from ESS, AKM, and Wolfson. There are now more than two companies with custom digital filters (some using DSP chips and others using FPGAs, but they still comprise the minority in this area (Ayre, dCS, Chord, PS Audio in their latest products designed by Ted Smith, Aesthetix, Schiit in their top model(s?), Auralic just recently announced at least one model, and a few more I'm sure I'm forgetting).
Slow-rolloff digital filters didn't hit the mainstream until Pioneer adopted them in the early '90s, calling this feature "LegatoLink". Pioneer's purchasing volume was large enough that Burr-Brown included a slow-rolloff option in their digital filters. Ayre was the first (to the best of my knowledge) to use this Burr-Brown digital filter in a high-end product, the D-1 DVD/CD player, released in the late '90s. The digital filter could be selected by the user via a toggle switch on the rear panel labeled "Listen/Measure". The slow-rolloff ("Listen") position yields superior performance in the time domain (less ringing - or as MQA calls it, "time blur"). This comes at a cost in the frequency domain as there is a -3dB rolloff from 15kHz to 20kHz. (This is all part of the "no free lunch" rule.) If one reads the reviews, the "Listen" position is almost universally favored. The slight rolloff in the very top half-octave is practically meaningless given the variations in frequency response due to loudspeakers, headphones, rooms, and individual hearing.
Three decades after Wadia, MQA seems to have discovered the benefits of using a slow-rolloff filter. At the 192kHz sampling rate, the MQA process has a -3dB point at around 33 kHz - roughly 1/3 of the 96kHz Nyquist frequency used by typical "brickwall" digital filters. I suspect this is the reason that some like the sound of MQA - the MQA-mandated slow-rolloff digital filter sounds "better" than the sharper digital filter found in most DACs.
But even this is up for debate. Take a look at JA's recent review of the Chord DAVE DAC, priced over $10,000. It has a brickwall filter with tens of thousands of taps, and the most pre- and post-ringing of any digital product JA has measured. Yet he found the sound of that unit to be excellent also. It would seem that there may be some other factors at play that are as yet unidentified.
Thank you for your detailed over view of DAC filter, design, and architecture!
And yes, there are many factors at play with respect to listener preference.
"If one believes that bit depth is the most important parameter, this would not be much of an improvement over a standard Redbook (44/16) file. On the other hand if one believes that the sampling rate is more important (as do I, as it allows for much gentler, less destructive filters), then there is a benefit."
What would we do without you 'smart guys' to explain this stuff in a manner that even us lame audiophiles can understand?
And while one could argue (and MANY here DO) that the resulting file is NOT indistinguishable from the original Hi Rez master from which it was derived, that's hardly the point if the only other alternative to the consumer is a 16/44.1 FLAC version of the same recording.
And agree it's great to have the option of streaming 'hi rez' from QOBUZ in the form of their Sublime+, should it ever make it to our shores.
Again, thanks for the most eloquent of posts on a subject which has been much abused on these boards.
The question that up for grabs is whether MQA's solution is the best for your needs. While it allows for a "gentle" digital filter on the playback side, it requires that one start with a true high-res file and then discards a significant part of that resolution to maintain a constant streaming bandwidth, and also requires the purchase of new hardware.
When streaming true high-res files (as with Qobuz Sublime+) one can have their cake and eat it too. The full 24-bit resolution is maintained, one doesn't need any new hardware, and if they like the sound of a "gentle" digital filter, they can buy one of many DACs that provides that option.
One thing to note is that the digital filter has a noticeable impact on the overall sound quality of any digital product - along with a couple of dozen other factors. There are many companies that have expended considerable resources to develop their own custom digital filters that they believe outperform the one mandated by MQA. Obviously you as the end customer have the final say as to which you prefer.
" Sublime+, should it ever make it to our shores."
If ever indeed. Qobuz' expansions plans this year involve adding Italy (done) and Spain and Poland later in the year. That will give them a reach to 7 EU member states by the year end. Only another 21 to go! Then there is the rest of Europe. Of course if their contracts with suppliers defines "EU Member States" for the contries to be served and not country by country, which I suspect could well be the case as this is typical these days in our single market, then not only will the USA not be supplied but I will lose my UK connection when we Brexit.
NB: the "free sample" playlist from Qobuz for Sublime + appears to offer mainly 16/44.1 files which is weird in the circumstances. There is one 24/44.1 file. I have queried this with them and await their response. I thought that my DAC's display may have been misbehaving but, no, 16/44.1 is shown in the Qobuz app itself for these tracks.
Mqa is not getting rid of the bottom 8 bits. They are actual encoding additional information into those bits and then recovery that content in the decoding or as they call it the unfolding process. Your way gets rid of those bits but does not add any additional info
It seems you are asserting the low order bits *are not* replaced as Charles Hansen says.
If that *is* the case could you please share what you know on the topic ... which would presumably be that origami folding (encoding) of 2FS and 4FS audio info into the low order bits is done in a manner that allows (on decoding) recovery of LSB bits to original values (in addition to the 2FS and 4FS info)
If that *is not* the case then please re-read my post and the Hansen posts referenced within.
I don't understand what you are saying. I am saying that the lower 8 bits of the 24 bit MQA file have additional content stored in those bottom 8 bits.
Oh thanks ... and since I can't make it any simpler for you let's just leave it at that.
Charles Hanson above said the exact same thing I said. Maybe it's you that has no idea what your'e talking about
Post a Message!
This post is made possible by the generous support of people like you and our sponsors: