@joel got it slightly wrong I think, and @MusicEar got it more wrong. MQA is completely lossless through the first “fold”. It’s goal is to keep all the data–not that pertains to human hearing, but that pertains to recorded music. By examining a lot of different music files, they’ve determined that musical information–whether we can hear it or not–goes up at most to about 60kHz. Most recordings don’t go nearly that high, partly because the instruments don’t generate harmonics that high and partly because engineers use microphones that don’t go that high.
So, MQA is lossless well beyond human hearing (in the frequency sense), up to frequencies that encompass nearly all musical information–that is, information that does not correspond to noise.
Yes but it seems that the definition that many people (including hiresaudio) want to apply to lossless is that it preserves every bit in the 192/24 or whatever master not matter if it contains any relevant musical information or not.
By that definition MQA is not lossless. By your definition it is
Yes.
And that is why I argue that the debate about the word is meaningless.
We know exactly what MQA is (in this respect, at least).
So using “lossy” as an epithet is not useful.
The only meaningful question is if this is a reasonable engineering decision, if it is fit-for-purpose.
What’s interesting to me in the MQA context is that it’s one long journey toward focusing more on the time domain. Of course this goes back beyond apodizing filters–to the earliest days of digital audio and “jitter,” which in those days, if memory serves, was thought to add an unpleasant anharmonic overtone to sounds. That was its supposed fault. People fixed a lot of jitter, then apodizing filters came along to in a sense correct existing timing errors. PS Audio is another company that hasn’t totally embraced MQA (although they intend to do a partial unfolding in their Bridge with a firmware update soon)–and yet if you watch videos with Ted Smith, the DirectStream DAC designer, he too is focused mainly on time-domain integrity.
What MQA does–if it does what it says it does–is to take this sort of thinking to a natural endpoint: MQA appears to even trade some aliasing for improved (and according to the patent applications quite outstanding) impulse response. I’m not discussing the “origami” thing here because I find that less interesting. As others have said, it’s not obvious that the compression is necessary.
The premise is that trading a little frequency-domain aliasing for a substantial improvement in impulse response is worth trying, just to see how it sounds. Seems reasonable to me. That’s the test–how it sounds.
Yes I have had some of the same thoughts when reading about apodizing filters but I didn’t connect that it went back even further.
And also I agree that for the pure audiophile the compression/origami is less interesting. It does nothing for sound quality, but it does mean that since I already have a Tidal subscription I can get access to high resolution music without having to pay for them for the third time in my case (LP and CD) but maybe the fourth or fifth for others if they went for SACD and HighRes PCM as well.
I assume there are economics here that I don’t understand. Apparently the compression DOES matter to someone; maybe it matters more to the streamer (Tidal, for instance) than it does to the stream-ee (me). As others have pointed out, the bitrate of, say, 24/96 is modest compared to HD Netflix. And it certainly is nice for downloads: An MQA file is (as others have noted) much faster than a DXD file.
But, yes, it’s the time-domain bit that’s most interesting. Consider it an experiment. Let’s listen and decide whether it improves the sound. That is, after all, what matters most.
@Jim_Austin, what @joel has mentioned is not wrong, it is based on psycho-acoustics principle using human hearing modelling in order to cut down the number information so it can take up less bandwidth to stream. That’s the whole idea behind it. This was more than two years back, but now internet speed has improved to a point, people are streaming 720/1080p on their phones, at such speed it is more than enough to stream FLAC at 96kHz and beyond without any form of compression.
It can however, achieve some degree of ‘lossless’ if the there’s no channel overload given at the point of time. We can’t guarantee for sure there’s no channel overload given the particular piece of music contents, it depends… The music content is a piece of complex variable information and we can’t simply draw conclusion like this. When we meant ‘lossless’ is all information is recoverable from the original recording.
Apodization means “removes the foot” and has a long history as a technical term in optics and astronomy prior to it’s adoption in audio. At it’s widest it means anything that changes the shape of a mathematical function, which would extend to every filter ever devised. Accordingly it is necessary to know what filters the user of the word regards as apodizing before the meaning of any statement can be determined.
As I have described, some people limit use of apodising to a particular type of function changing filter, rather than applying the term as it has been used historically in other contexts. The result is pure confusion and before taking any claims that a particular filter is or is not apodising seriously we need to know what the user means when they use that term. It has become an unhelpful word to use and in my view we will all be better placed not to use it.
Perhaps, but I believe usage is fairly consistent in audio; the Ayre white paper linked to above says it quite well:
“In Peter Craven’s 2004 AES paper, he proposed that the playback DAC should include a digital filter that had a corner frequency below the half-sample rate of 22,050 Hz. This would filter out any ringing (pre- and post-) that was introduced during the recording process and thereafter embedded on the disc itself. He named this an “apodizing” filter.”
An apodizing filter “fixes” ringing introduced by previous filters.
I admit I’m quite confused by this; maybe you, and @joel, are talking about the generic idea of lossy compression, mp3-style, and assuming it applies to MQA as well? But as I’ve said, up to audio (not sampling) frequencies of at least 44.1 kHz, and often/usually 48kHz–frequencies that completely capture all musical information present in MOST recorded music, whether we can hear it or not–MQA is completely lossless, like FLAC, not like MP3. Except for the first few kHz above that level–a small amount of recorded music has musical information up to about 60kHz–it’s only noise, not music, that’s lossily compressed in MQA.
I base my assessment on extensive reading; I’ve studied the source documents on MQA, including the patent applications, and most of the measurements I’m aware of that I’ve seen online. It’s possible that you are relying on information I haven’t seen. If so, please cite your source. Thanks.
MQA is only lossy in a technical sense (not able to fully recover mathematically from the original source) that it doesn’t encode the entire rectangular Shannon space… Most of that Shannon space up to 24/192 and above is literally a waste of space, so there is little or no information in this discarded region (If it does, it will be buried under the noise floor) which is relevant to human hearing (this is psycho-acoustics human modelling principle). Thus encoding information can be drastically reduced to flat bit-rate ~1.5Mbps across sampling frequency from 44.1kHz right up to 384kHz and beyond. Thus this save a lot of bandwidth!
There are of course certain conditions HF (high frequency components, above 20kHz) that the touch-up channel can over load, and when this happens it will revert to lossy. Strictly in technical sense its final output resolution can never achieve a true ‘24-bit’ like in FLAC.
So does it really matters? The answer is No! Most modern ADCs and DACs are unlikely to achieve true 24-bit resolution in practice. Does technically lossy in this case which happens in the high frequency above 20kHz really matters? The answer is No! we simply not able to hear information that is near to noise floor or those buried in the noise floor.
Hmm, now I’m really confused, since we now seem to agree. Initially I was addressing this comment:
which seems wrong to me, or mostly wrong, since MQA is (the proprietors claim) lossless ALMOST through the whole range where there is any musical information at all–or, as you say, the whole range where it isn’t buried under the noise floor, which is saying much the same thing. Above that frequency is effectively only noise, and that is what’s being lossily compressed.
So how is it that we now seem to agree? Apologies if I simply misunderstood you.
The triangular shows in two domains, frequency and spectral level, for frequency up to ~60kHz is encoded and for spectral level, anything below -120dB in the audio band is not encoded, this increases up to well below the -96dB of a 16-bit CD resolution at ~60kHz. So what about anything above 60kHz? It is discarded. Spectral level below -120dB in the audio band It is discarded. Just below -96dB at around ~60kHz, it is discarded. In a classic PCM, it just encoded everything beyond the limits of the triangle all the way the maximum nyquist frequency, thus the bit-rate has to increase with the function of the sampling frequency.
What are we discussing here regarding MQA?
Lossless not lossless. Different filters. Whos right, who is wrong tecnically. This is absurd.
The only thing that matters is, do you like what you hear when you listen to the music on a hardwere MQA licensed DAC or not. If not don’t listen to MQA and stop arguing that others are wrong when they appretiate their music via MQA.
FYI, I do listen to MQA via the Tidal app and it sounds better than the CD Redbook. I’m trying to dive down on the technical aspects of MQA, don’t you really want to know?
How is it “absurd” to discuss how the systems we listen to work?
Yes, how it sounds is important to all (I’d think), but how it works is just as important to many (should be “everyone”, but some people are incurious).
OK, I THINK I see what you’re arguing here. I’ve read similar claims elsewhere. Some aspects of those claims appear to contradict what MQA folks have said publicly about their technology. That makes it difficult to know what to believe. I’ll just note that MQA has provisions for 2x ad 4x transmission rates that would take their lossless compression up to 192kHz/DXD respectively. I’ve said before that I find the compression to be the least interesting aspect of MQA; it seems like something they did to show it off, because it’s clever, and not because it was strictly necessary. Then again, none of this troubles me very much; if the problem you’re describing is real, and if it causes audible problems with complex material, that should reveal itself in time. I remain most interested in MQA’s time-domain focus.
I don’t dismiss theoretical/technical analysis as worthless, as some do. But I’ve seen no objections raised that I find troubling. I’m convinced enough by the technical stuff to believe that it all comes down to what it sounds like.
