Most-Read Opinion Articles (Last 365 Days)
- 2017-01-01 - Our 2016 Products of the Year -- One-Sentence Summaries
- 2016-11-01 - The Best of the 2016 Tokyo International Audio Show
- 2017-02-01 - The Best of the Worst CES in Decades: 2017
- 2017-05-01 - Is It Finally Time for Active Loudspeakers?
- 2017-04-01 - MQA One Year Later -- Suddenly, More Questions
- 2017-06-01 - The Best of High End 2017
- 2016-12-01 - Warsaw: The Best of Audio Video Show 2016
- 2017-03-01 - Sonus Faber and Vivid Audio -- New Ways to Launch New Products in the New Year After a Dismal CES
- 2017-09-01 - A Poor Man Can Only Afford to Buy the Best -- Three Tips for How to Avoid a Bad Buy
- 2017-10-01 - Mismatched Masters and False Frequencies -- Is MQA Better, Worse, or Just Different?
- Written by Doug Schneider Doug Schneider
- Category: Monthly Column Monthly Column
- Created: 01 October 2017 01 October 2017
Master Quality Authenticated (MQA) digital-file playback has been a controversial topic in the hi-fi press ever since it was first announced, in December 2014. Usually, the controversy around a new audio technology cools down after a while. But with MQA, it just gets hotter, as recent tests by SoundStage! and other journalists and manufacturers are raising more questions that have so far received few answers.
Originally developed by Meridian Audio, of the UK, MQA has since been spun off as MQA Ltd., a separate British corporate entity. Bob Stuart, who founded Meridian, has been the face of MQA since the beginning, and has left Meridian to work only with MQA Ltd.
MQA is more than just a codec. Its multiple aspects include a proprietary digital filter, and a lossy data-compression algorithm that shrinks any PCM-based high-resolution digital music file into a file the size of a 24-bit/44.1kHz or 24/48 PCM file of the same duration. In full MQA playback, the system is claimed to “unfold” the audio data until it has its original bit depth and sampling frequency. (High resolution means any music file with a bit depth and/or sampling frequency higher than the CD standard of 16/44.1. In lossy compression, data are discarded that cannot be restored when the file is decompressed.) The main benefit is in streaming: smaller files reduce the amount of data flowing through a network. In fact, the month MQA was announced, John Atkinson, editor of Stereophile, wrote “I’ve Heard the Future of Streaming: Meridian’s MQA.” However, as time goes on, and networks get faster and the amount of data that they can easily process grows, the sizes of files become less important.
But MQA wants their codec to be used for more than streaming, and it’s Robert Harley, editor of The Abso!ute Sound, who seems to have become the strongest proponent in the audiophile press of pushing MQA to be used for all digital playback in the audiophile world. In 2015, in “Beyond High-Resolution,” Harley raised many eyebrows when he wrote: “Master Quality Authenticated is an innovative and sophisticated new method of encoding digital audio that simultaneously improves fidelity and lowers the bit rate.” In the September 2017 issue of TAS, Harley doubled down on his praise for MQA in “Let the Revolution Begin: MQA and the Overthrow of 20th Century Audio.” In that article he describes MQA as representing a “paradigm shift,” and states: “It’s this new framework that some can’t comprehend, along with a reluctance to abandon long-held beliefs in certain ‘scientific principles.’ But I suppose we should cut the critics some slack. After all, if Lord Kelvin could have been so wrong about the state of physics in 1900, it is easy to understand how a few audiophiles could be so mistaken about MQA.”
Obviously, Harley thinks highly of MQA -- but is he right?
That MQA lowers the bitrate of a high-resolution file is easy to accept -- MP3 and many other lossy codecs shrink file sizes in this way -- but the “improves fidelity” part stopped me and others in our tracks. Put the two together and they mean that fewer data can bring us closer to the sound of the original recording, which seems counterintuitive, to say the least. Merriam-Webster’s Collegiate Dictionary, 11th Edition, defines high-fidelity as “the reproduction of an effect (as sound or an image) that is very faithful to the original.” But how can MQA improve the original by changing it -- by even discarding data?
Lossy though MQA is, its supposed higher fidelity has since been championed by other writers, and enough consumers that if you Google “MQA better than the original,” you’ll find pages of links you can click to. Search for yourself.
But does MQA produce better sound than the original file, or simply a different sound -- one that those who like it prefer? Note that I said prefer. Sound quality is subjective; just as some people prefer the sound of tubed to that of solid-state electronics, or the sound of vinyl to CD, is that why some prefer MQA? I’m not sure yet, but having closely researched MQA for well over a year now, I’ve found two possible reasons why MQA playback could sound different.
There have been many comparisons of MQA files and other music files, but from what I can tell, it’s questionable whether the files being compared were made from the same master file -- something I first brought up in April 2016, in “Myriad Questions about MQA.” Using identical masters is an imperative first step when doing valid comparisons.
For example, in “MQA’s Sound Convinces Hardened Showgoers,” a January 2016 article on Stereophile.com, Jason Victor Serinus describes his impressions of a demo conducted by Bob Stuart, and quotes remarks by fellow Stereophile writer Michael Fremer about a comparison of Keith Jarrett’s The Köln Concert, originally released in 1975. Jason wrote: “Lucky for us, Stuart had been able to get his hands on a 96kHz transfer of ECM’s analog master tape.” Yet Fremer’s comments about the comparison were: “The CD of the recording has an unfocused, diffuse image of a piano hanging in space, with the room reverb mixed in and confusing the picture.” Does this mean that the new 96kHz master was used for the MQA-encoded file and that it was pitted against a CD pressed some time ago? If so, it was definitely not two versions of the same master that were being compared, and perhaps not even the same mix. I have no reason to doubt Fremer’s description of what he heard, but what, exactly, he was listening to is not clear. Unless we know that the CD was produced from the same 96kHz transfer as the MQA file, the comparison isn’t valid.
More recently, I’ve seen Donald Fagen’s The Nightfly used for comparison, as in John H. Darko’s “MQA, Donald Fagen & PS Audio,” on DigitalAudioReview.net. Originally released in 1982, this album has an interesting history in that it was one of the first digital recordings made -- they used 3M machines running at a sampling frequency of 50kHz, not CD’s 44.1kHz, or the 96 and 192kHz common today. Since the 3M machines had no digital outputs, anything recorded on them had to pass through their analog outputs, meaning they first had to be converted to analog by the 3M’s internal DACs. This meant that subsequent digital releases of that album, including CDs, had to be redigitized by an ADC -- another conversion.
In his article, Darko states that he got the MQA and 16/44.1 non-MQA files of The Nightfly that he compared from the Tidal streaming service. I was eager to hear the differences. Although I don’t have an MQA-compatible DAC, MQA-processed files can be played by non-MQA DACs, and some have reported that they sound better than the non-MQA versions. I cued up both and, without looking at which one I was playing (making this comparison pseudo-blind), I switched back and forth between them. Considering that I couldn’t take advantage of full MQA unfolding, I expected to hear, at most, only subtle differences, along the lines of what MQA promises in its statements.
Instead I heard obvious differences, beginning with the first track. The standard 16/44.1 version of “I.G.Y.” sounded louder, with better macrodynamics, and stronger, deeper bass than the MQA version. A loss for MQA? Just wait. With track 2, “Green Flower Street,” the playback levels seemed about the same, as did the sound, though I did feel that the MQA track was subtly softer in the highs. A win for MQA? Or a draw? No. The fact is, neither won.
When I played track 2, I noticed something else: It went on far longer than the non-MQA version. When I checked the track times, I found that the MQA version of “Green Flower Street” is seven seconds longer. I then checked the rest of the album, and discovered that four of the seven other tracks differ in length by a second or two. Clearly, these two versions were not made from the same masters. Therefore, anyone who’s comparing them isn’t comparing only MQA to 16/44.1 playback; they’re also comparing the differences between two masters -- and given all the conversions the music on this album has gone through over the years, you’d need a detective to know which one is closer to the original source files.
My experience with The Nightfly is important. Tidal streams MQA and non-MQA versions of many albums, and many audiophiles automatically think they can be directly compared to hear if the MQA versions are better or worse. All I can say is that any validity such comparisons might have will heavily depend on whether identical masters were used, and Tidal doesn’t tell you if they have been. (See the “MQA is Vaporware” thread on ComputerAudiophile.com. This link takes you to p. 143, where The Nightfly is discussed.)
That’s not to say that there have been no comparisons of matched masters. The 2L label, of Norway, has made available free downloads of some MQA tracks and their non-MQA hi-rez counterparts. I have no reason to doubt 2L when they state that the MQAs were derived from the same hi-rez files offered for download. I’ve listened to some of the tracks, on non-MQA DACs, and haven’t heard any differences that I could reliably identify under blind conditions -- certainly not like what I heard with The Nightfly. Stereophile’s editor, John Atkinson, has had some of his own recordings converted to MQA, which he wrote about in “Listening to MQA”; however, those recordings haven’t been made publicly available, so unfortunately we can’t hear them. He describes his own pseudo-blind tests with the tracks, in which he was able to identify MQA vs. non-MQA four times out of seven. Likewise, Peter McGrath, of Wilson Audio Specialties, has had his own recordings MQA’d, and these have been used in MQA demos -- but they’re not publicly available, either.
Unfortunately, other than the examples above, MQA itself has not been forthcoming with sets of MQA and non-MQA versions of the same recordings verified as having been made from the same master files. This is why I believe that there haven’t been enough proper comparisons of MQA and other formats, and that any differences people are hearing might be attributable to hearing different master files. The lack of proper comparisons is echoed by Charles Hansen, founder and chief designer of Ayre Acoustics, who, on August 23, 2017, in a lengthy post in a highly contentious thread about MQA on Audio Asylum in which I participated, wrote that, as for true apples-to-apples comparisons of MQA, “it’s never been done.”
MQA is tight-lipped about the technical details of their system. Even their patents are hard to glean much information from. What they do admit is that the MQA process creates “aliasing,” which, as Eric Tarr says in “Digital Audio 101: Aliasing Explained,” a December 2014 article on ProAudioFiles.com, is “one of the more complex concepts of digital audio.” Complex or not, it’s worthwhile to understand at least a little bit about aliasing, to get a better handle on what people might be hearing from MQA playback.
Aliasing can occur on the digital recording and/or playback side. Basically, it’s “phantom” sound caused by a digital audio system as it attempts to record or process audio frequencies higher than the Nyquist frequency, which is half the sampling frequency of the format used, and defines the very highest frequency that that format can record, store, and reproduce. For example, the Nyquist frequency of the CD standard is 22.05kHz, which is precisely half the standard’s sampling rate of 44.1kHz, and 2.05kHz above the upper limit of human hearing: 20kHz. Digital audio systems prevent aliasing through the use of low-pass filters that remove high frequencies that exceed the Nyquist frequency.
In his article, Tarr summarized what aliasing is on the playback side:
A conventional D-to-A converter for audio will only reproduce signals within a specific frequency range that is determined by the sampling rate.
If the D-to-A converter is fed any recorded frequencies outside of this range, they are interpreted by the converter and mapped to frequencies within this range. This is aliasing -- when one frequency is coded as a different frequency.
When one frequency is coded as another through aliasing, that’s a form of distortion that is inherently amusical: that is, it is sound added to playback that was not part of the original recording, or the original musical event recorded.
On the recording side, frequencies higher than the Nyquist frequency can be mapped as lower frequencies and end up in the recording, if they’re not filtered out before the analog-to-digital conversion stage. Conversely, on the playback side, lower-range frequencies can get mapped into higher-frequency ranges if a filter isn’t used to eliminate these unwanted, higher false frequencies, which will wind up being transmitted through your system. This is the reason that, even in the early days of CD, they used “brickwall” (i.e., very steep) filters to truncate the output abruptly at the CD standard’s Nyquist frequency of 22.05kHz: precisely half of the format’s 44.1kHz sampling frequency. In a nutshell, the filter was there to make sure no frequency higher than 22.05kHz made its way into the playback system. For the newer, higher-resolution files, which have higher sampling frequencies (e.g., 96 and 192kHz), the filters used are higher in frequency.
MQA doesn’t use a brickwall filter. Instead, it has a filter with a very slow rolloff that, MQA says, optimizes the time-domain accuracy, for more accurate reproduction of transients. As with most things in audio, this approach requires tradeoffs. The tradeoff with a slow-rolloff filter like MQA’s is that aliasing can leak through, which is why, in some articles and on some forums, MQA has been described as having a “leaky filter.” This should be no surprise; in MQA patent #WO2015189533 A1, “Digital Encapsulation of Audio Signals,” they state: “We accept that aliasing may take place and are proposing to balance aliasing against ‘time-smear’ of transients due to the lengthening of the system’s impulse response caused by filtering.”
Is the tradeoff worthwhile? It depends on whom you ask. According to John Siau, VP and director of engineering at Benchmark Media Systems, it’s not. He not only disputes MQA’s claim of improving time-domain accuracy, he says it makes it worse, as he wrote in “Is MQA DOA?”: “With MQA’s adaptive filtering, transients are subjected to a distortion mechanism known as aliasing. This aliasing causes tonal shifts of high-frequency content and it causes small errors in the timing of fast transients.”
I’m not in a position to say who is right regarding the issue of timing, but I will say that, from what I’ve seen, neither MQA nor anyone has offered proof that their system improves timing accuracy. However, there is proof that the aliasing is real.
I was impressed to see Paul Miller begin to address this topic in the July 2017 issue of Hi-Fi News & Record Review, which he edits, and for which he measures electronic audio components. On p.103, in “More Questions Answered?” (which he gave me permission to excerpt here), Miller plots the outputs of MQA-encoded files through MQA and standard filters. The results are enlightening.
Laura Metcalf and Matei Varga’s First Day is purported to be a 24/192 recording, which means that the musical information can be captured up to the Nyquist frequency of 96kHz, or half the sampling rate. Miller’s chart shows that by about 48kHz, any musical information has sunk into the noise floor, so it’s gone. This means that a 24/96 recording would have sufficed, as that can capture up to 48kHz. However, because it’s a 192kHz recording, the range of 48-96kHz has to be filled in, and the MQA process appears to do that -- not with information from the recording itself, or even silence, but with aliasing. As Miller’s chart of the MQA output shows, above 48kHz, the signal begins to rise as the frequency increases, until it reaches 20dB by about 80kHz, then tapers off slowly into the noise floor just shy of 100kHz. Miller points out that the signal above 48kHz isn’t musical information but aliasing, the result of lower frequencies being mapped into higher ones that are seeping through the MQA filter.
Miller’s measurements of a 24/192 version of the title track of Metallica’s “Enter Sandman” show that musical information is once again gone by 48kHz, but above that frequency, the aliasing from MQA playback is more pronounced -- beginning at about 48kHz, the signal slowly rises until about 75kHz; then it rises quickly to about 40dB at 80kHz before descending again, to trail off into the noise floor at about 96kHz. It then jumps up again at 96kHz, then cuts off, since the graph goes only to 100kHz. Again, the frequencies above 48kHz all appear to be the result of aliasing -- false frequencies that weren’t in the original recording and shouldn’t be there. Looking at this graph in particular, I was shocked at how high in level the aliasing is.
Some might feel that these aliased frequencies are so far above the audioband (i.e., 20Hz-20kHz) that we can’t hear them anyway, and so they should be benign. However, when using an audio component not designed to reproduce ultrasonic frequencies, strong content in the ultrasonic range -- whether aliasing or actual musical content -- may produce intermodulation tones that make their way into the audioband, where they’re readily audible. For example, a 40kHz recorded tone adjacent to an alias at 42kHz can produce a difference tone at 2kHz, which is right in the human ear’s “sweet spot.” Until we know more about what effects it can have, I’m not sure I want all that high-frequency energy traveling through my preamp, amp, and speakers -- could it damage them, particularly the tweeters?
The main question this article poses is: Is what some people report hearing when they listen to MQA a result of this aliasing? I can’t be sure right now, but I’d say that more research is needed -- something Miller seems poised to do, and something that we plan to do at the SoundStage! Network.
Also interesting is Miller’s graph showing plots from 24/96 PCM and MQA files of Jarrett’s The Köln Concert -- which I have to presume are identical copies of the same files mentioned in Jason Victor Serinus’s article. The plots of the files aren’t identical, but they’re similar, and with no aliasing -- even the MQA one. This indicates to me that the MQA system works best with 24/96 source material. The similarity makes me wonder what the differences in sound might be between those two files -- if there are any differences at all. Miller doesn’t comment on that in his article, but it’s a comparison I’d like him to make.
What are we hearing?
I have no doubt that the differences people are hearing from MQA are real. But what is causing those differences? If the cause is different masters being used, then any difference, good or bad, shouldn’t be attributed to MQA -- it’s simply a difference between two master files. It’s not unlike how a new mastering of a recording released on CD can sound markedly better or worse than, or just different from, an earlier mastering.
Insofar as aliasing goes, then, it’s valid to point to the file formats. MQA Ltd. seems content with so much aliasing, but most digital engineers will tell you that aliasing is to be avoided because it pollutes the audio signal with frequencies not in the original recording -- i.e., noise. Such noise is, by definition, distortion, and makes the result less faithful -- i.e., of lower fidelity -- to the source. Is the frequency-related distortion of aliasing what some are hearing in MQA files? Obviously, more work needs to be done in this regard; Miller’s measurements are telling, but only the tip of the iceberg.
As for Robert Harley’s latest comments in The Abso!ute Sound: Perhaps I’m one of those critics who doesn’t get it -- or maybe he doesn’t get it. I won’t say that he’s right or wrong, but I think he’s jumped the gun -- there’s so much about MQA that remains unanswered and needs to be explored before anyone can declare its results to be better than, worse than, or simply different from the original file, and it’s not likely to be resolved anytime soon. Stay tuned . . .
. . . Doug Schneider