MQA creator Bob Stuart answers questions.

[RayDunzl]

Please clarify. Are you saying that sampling at 1 Hz can provide the same timing precision as sampling at 1 MHz?

Maybe yes, for a tone below 0.5Hz (1/2 sample rate)

 

[Cosmik]

Montgomery's law. When an ASR poster has to be pointed at Montgomery's video the quality of the thread will trend towards zero.

The opposite, surely? Until everyone understands the beauty of digital audio, sensible discussion is impossible.

 

[Blumlein 88]

Maybe yes, for a tone below 0.5Hz (1/2 sample rate)

You'll have to use one of those rotary woofers for this. They work below 1 hz.
https://www.soundandvision.com/content/eminent-technology-trw-17-rotary-subwoofer

 

[LTig]

[..]
You are using a scope and a frozen graph of the signal amplitude. The hearing system doesn't have such luxury. It can't "see" even instantaneous value of a signal, leave alone a high-resolution recording of the signal over time.

This has nothing to do with the hearing system. The task of a transparent audio chain is to reproduce the sound captured by the microphones by the speakers in your room (well, ignoring mixing and mastering of course). You said that 44/16 cannot reproduce transients faithfully due to limited timing resolution. I showed a proof were you can see with a scope what you cannot see or hear with your ears, namely two signals shifted by a delay about 22 times shorter than you stated is possible (in fact the timing resolution is many magnitudes better than what you stated, as has been demonstrated elsewhere). And what the scope measures (voltage) translate accordingly into wave pressure at the speaker.

BTW, you can do this test with any signal you like.

 

[Cosmik]

How many people, to this day, listen to digital audio under a burden of doubt due to half understanding how it works?

I think there must be a whole load of people who assume that what they're hearing is basically rubbish, but by throwing enough bits and hertz at it it can be made to sound good enough for the masses.

But they're thinking that down at the lower volume levels it really sounds like a Commodore 64 but is quiet enough to hide it. And they're thinking that at high frequencies it's a stairstep timing disaster but close enough to being ultrasonic that it's masked.

If only they realised that it's perfect they might begin to enjoy it more.

 

[SIY]

How many people, to this day, listen to digital audio under a burden of doubt due to half understanding how it works?

Very, very few. They're just disproportionately noisy. The vast majority don't understand how it works and don't care, they just want to hear the music.

 

[dc655321]

Until everyone understands the beauty of digital audio, sensible discussion is impossible.

That needs to be put on a t-shirt.

 

[Sergei]

If it's not about the money then what is it about.
De-blurring was a process that a one time was claimed to be separate from the compression scheme. OK, so offer de-blurring by itself and if people find it of value they'll buy it.

Agreed.

The compression is a answer to a non-existent problem. The small amount of bandwidth it saves when balanced against the data losses over a PCM 24/96 stream doesn't compute, it's not worth it. Most people say it sounds different but is that better or worse, maybe it's the compression losses heard?

I agree with you considering American market. Globally, the situation is quite different. In a number of developing countries, people only have access to Internet via cellular, and there are no affordable unlimited data plans. What our family pays for cellular and cable data access exceeds average Indonesian's income, as of 2017. Compression is still very important for those markets.

The record companies have jumped on board cause it gives them what they wanted from DRM all along, no public access to a uncompromised digital copy of the original master tape. MQA is not Master Quality Authenticated, it's something else, a bastardized version of the original master done with the excuse of bandwidth savings for streaminng, a savings not needed in 2014 and all but laughable today when compared against the video streaming going on.

Video streaming is going on in developed countries. What's going on in many developing countries is video pirating. I've been to Indonesia recently. On one of the nights I happened to be deep in a residential area of one of their major cities, far from tourist traffic. I had a delicious dinner for $1.20. Yes, for one dollar twenty cents. I just priced dinner with exact same dish and similar drink at a restaurant here in Palo Alto, California. $17.

The 14x price factor is not the limit. Indonesia is a relatively prosperous country, with abundant natural resources, educated population, and plenty of local businesses. In a remote location in Vietnam, I visited a family with monthly cash flow three times less than what our family pays for cellular plan. And again, Vietnam is a relatively prosperous country. People in many developing countries, especially in Africa, have much less free cash available.

It's all about the money, and cutting off public access to bit perfect copies of the original masters
Just say NO.

I agree that these objectives are in play. But they are not the only ones. I'm saying NO to MQA for now, but not for the reason of it being an "evil scheme". For me, the reason is that only about 0.2% of Tidal's catalog is available in MQA. With the current conversion rate, they'll get to the full catalog in MQA in about a thousand years. I don't have time to wait for that long.

This has been a sad corruption of the truth, again done in the name of the almighty dollar.
If CD 16/44 is standard res, then high res has to be something greater. But not just on the distribution end, but from the mic forward. In short a high res offering should have started as a digital recording at maybe 24/96 or better and never sampled to less than that.

Agreed.

You can't take a master analog tape done in 1960, transfer it to 24/192 and call it a "high resolution recording". That's a scam put forth simply in the name of $, once again being able to resell all the old catalogs of music with a promise of better SQ. There's nothing on those tapes that can't be captured at Red Book.

I disagree. A studio tape can capture the frequency range of 10Hz to 20KHz without anti-aliasing filter that CD requires. Dynamic range is sufficient for most music, which rarely needs more than 30 dB. And timing of the transients is virtually perfect. This timing can't be perfect on a CD.

Paraphrasing you, "a sad corruption of the truth" is that CD was advertised for so many decades as a perfect music carrier, whereas in fact it falls short of the PCM parameters that as we now understand are essential: at least 20 bits and 120 KHz sampling rate.

Similarly vinyl. Consider, for instance, that it is practically possible to record a signal of up to 122 KHz at 33 1/3 RPM (http://www.positive-feedback.com/Issue2/mastering.htm). Sure, it is sonically unnecessary, yet it demonstrates that the timing of transients can be captured in a manner equivalent to 244 KHz PCM.

Once again the public gets screwed in the name of record labels profits.

Isn't it always the case I'd be happy if more things in life were cheaper, or free, yet life is stubbornly refuses to cooperate on this.

 

[SIY]

Paraphrasing you, "a sad corruption of the truth" is that CD was advertised for so many decades as a perfect music carrier, whereas in fact it falls short of the PCM parameters that as we now understand are essential: at least 20 bits and 120 KHz sampling rate.

Cite badly needed here.

 

[amirm]

Cite badly needed here.

The 20 bit part is defensible actually. Ironically it is from the seminal paper by Bob Stuart in J. of AES, Coding for High-Resolution Audio Systems: https://secure.aes.org/forum/pubs/journal/?elib=12986

Only the 20 bit line is below threshold of hearing at all frequencies.

WIth noise shaping this requirement can be relaxed substantially to 14 bits. But then it would be good to have room to park all that noise in ultrasonics and hence the requirement for something better than 44.1 kHz sampling.

 

[Sergei]

Google found a nice short article on this:

https://science-of-sound.net/2016/02/time-resolution-in-digital-audio/

"So for a worst case of a 10 Hz sine with -60dBFS magnitude quantized at 24 Bits resolution, we get into a range of around 4 microseconds resolution. Which is already pretty coarse, but you wouldn’t hear that sound anyway. For a more realistic 100 Hz sine at -20 dBFS we are in the range of 4 nanoseconds. By the way, the samplerate doesn’t even show up in the equations!"

It's useful to emphasize that last point: the error in reconstructing a band-limited analog input is due to quantization, not the sample rate. The timing error (I surmise) for a particular input signal would be the maximum amount you can shift the signal in time before the quantized representation would change.

The trouble with PCM is that it seems simple, but our intuitions about it (at least for those not well versed in DSP) can be very misleading. And there is double trouble when marketers are happy to exploit those incorrect intuitions and engineers who know better look the other way.

The article you referred to nicely demonstrates misconceptions about the real-life sounds that the hearing system is capable of perceiving. The article states, for instance: "In fact, in reality there is no such thing as a sound event, just continuous signals." But what about sonic boom phenomenon: sound pressure can rise from zero to pain threshold in just 20 microseconds? The hearing system perceives it perfectly well, with the onset timing resolution of about 5 microseconds. But 44.1 KHz PCM capture can miss the onset peak, and only start recording decaying mid-section.

Indeed, when we deal with a signal that can be represented as a sum of a limited number of sinusoids of constant frequency and amplitude, Fourier transform shines. But what about transforming a signal consisting of sinusoids that decay at differing rates? Or rapidly changing their frequencies in arbitrary ways? Especially when the number of such "wild" constituent sinusoids far exceeds the number of frequency bins? Real-life signals are often like that. "Oh, that's simple", Fourier aficionados say, "it is easy to show that such signals are not band-limited by the human hearing boundaries, and thus we shall band-limit them first". But when you band-limit the signal this way, it is no longer the original signal.

Coming back to MQA. As I understood, their algorithm first collects statistics about the whole signal, and finds the maximum frequency at which the maximum energy of the music signal equals the energy of noise floor. That's the cutoff frequency, the right vertex of the MQA triangle. MQA guys say that it typically is around 60 KHz. Then, roughly speaking, they sample the signal at a frequency twice of that cutoff, letting the frequencies above the cutoff alias (fold) back into the audible range. Since such folded frequencies are guaranteed to be below the noise floor, the combined noise floor will only rise to a level that will be still inaudible.

So, the MQA, as I understood it, trades increased noise floor in exchange for the original signal staying unscrambled by an antialiasing filter, and thus keeping its phase and onset timing purity, which, arguably, is perceptually important. The exact way the signal is further compressed is unclear, being proprietary. The compression may indeed introduce artifacts and distortions, similarly to any lossy codec. Still, I consider MQA to be just a modern perceptually-coding compressing codec, worthy of consideration, with its pluses and minuses in the context of particular music material. I don't particularly like the DRM aspect of it, and don't like its slow rate of adoption. Otherwise, it is just one more codec that works, why hate it?

 

[Sergei]

For band-limited signals, yes. The bit depth being equal, and for a signal band-limited to 0.5 Hz (your example, not mine!), the 1 MHz sample rate provides no more timing precision than the 1 Hz sample rate. The same is true for for a signal band-limited to 22050 Hz: a 768 kHz sample rate provides no more timing precision than a 44.1 kHz sample rate.

Theoretically, true. Practically, false. Fourier transform is a linear integral transform. Mammal hearing system is not linear.

Similarly, the overwhelming majority of filters used to band-limit a real-life signal before the digitization are based on the theory of LTI systems. Mammal hearing system is not time-invariant.

An additional problem with the classic Fourier line of reasoning is that it doesn't take into consideration the real-life quantization used by the hearing system, which is not only highly non-linear and time-dependent, but is also affected - at a given frequency - by signals of other frequencies, through well-documented phenomena of masking and spreading.

 

[JJB70]

I'm not sure that access to high-res audio is a particular priority for people in emerging economies for whom access to affordable broad band is not there. I'm not sure it's important to many people in the developed world, not just to people with little interest in enjoying music either as I know plenty of dedicated music lovers who are perfectly happy with 320k MP3, FLAC or CD.

 

[Sergei]

That's right. Sample rate does not determine the timing resolution. However, higher sample rates allow wider bandwidth and if you take advantage of that, then the higher frequencies will improving timing.

Your hearing system is not perceiving digital samples one by one. It perceives a continuous waveform post reconstruction filter. In that sense, the sample rate is immaterial.

The timing resolution depends on how much resolution (bit depth) you have relative to frequency of interest. The higher the frequency, the better precision you need to determine it accurately. The minimum timing is = 1/ (2*pi()*f*A*(2^b-1). "f" is the frequency of interest; A is the amplitude; and b is the number of bits. Note the absence of sampling rate in the formula.

For CD at 22.05 kHz and 16 bits this becomes 110 picoseconds which is far smaller than the sample rate of 44.1 kHz (22 microseconds).

I agree with your line of thinking. However, it is only applicable to the output signal. When we are considering the real-life-signal => ADC => DAC chain, we ought to understand that a real-life-signal, if Fourier-transformed, may have frequency components beyond the range of human hearing, yet still important for representing the onsets of transients with a temporal precision matching that of the human hearing system.

Human hearing system is not an a chain of LTI systems combined with Fourier transformer. It is cruder than those in many ways, yet it also is more sensitive to evolutionary important "sound events" than a purely LTI+Fourier system would be. Band-limiting a real-life-signal to the formal frequency range of human hearing robs the signal from some of the cues that the the hearing system is attuned to.

 

[SIY]

The 20 bit part is defensible actually.

I'd love to find mikes and mike preamps that would deliver this. I get the theoretical argument, but with actual program material, I can't see its relevance- and I note the absence of DBT data indicating that a 16/44 ADC/DAC is audible when inserted into a signal chain and volume controls aren't deliberately twiddled.

 

[SIY]

I agree with your line of thinking. However, it is only applicable to the output signal. When we are considering the real-life-signal => ADC => DAC chain, we ought to understand that a real-life-signal, if Fourier-transformed, may have frequency components beyond the range of human hearing, yet still important for representing the onsets of transients with a temporal precision matching that of the human hearing system.

Why do you repeat the nonsense of temporal precision? It's just out and out untrue, as has been patiently explained to you several times in several ways.

 

[Sergei]

You can do some comparisons from 2L.

http://www.2l.no/hires/

BTW, I've missed where they had an article with mastering engineers doing such comparisons. Can you show where it is?

Journal of the Audio Engineering Society, Volume 67 Number 5, 2019 May. Don't have a reference to a freely available online source.

 

[JJB70]

I still think the fundamental problem for a certain segment of the audio industry (including record labels) is that the Phillips and Sony engineers who developed the CD standard did their job too well, making it very difficult for anybody to dream up a new format which really moves audio on in any meaningful way. To me the big leap audio could make is to for multi-channel to find a market but that wouldn't need stuff like MQA either.

 

[SIY]

If you mean the Bob Stuart paper, it's here. Free download.

 

[Sergei]

Just a note regarding the alleged benefits for reducing bandwidth requirements and cost for streaming providers. Most people are happy with MP3 quality streaming. Those that arent (us) are a small subset of the market. That subset are willing to pay a premium for access to the uncompressed streams. As such the cost reducing argument holds no water. The customer will pay.

A first-hand perspective. I was long-intrigued by the seemingly irreconcilable gap between some people I respect raving about the Indonesian gamelan music, and my aversion to gamelan music that I heard on CDs, which always sounded to me like an incoherent mess. At some point, I went to Indonesia to hear it in its natural habitat. I was blown away!

It turns out that a typical gamelan ensemble emits a sound with such intricate spacial, spectral, and temporal structure that it can't be faithfully captured on a CD. My guess is the labor is so cheap over there that they kept adding performers until the complexity of the resulting sound became just below a typical human's ability to comprehend it

So, Indonesia is one of the examples of a large market where high-resolution recordings, delivered very inexpensively, could be quite successful. I'm not convinced at this time that MQA will be such format. Most likely, an open-source format, designed around the MQA patents, will ultimately fill the bill.