MQA is Vaporware

[Jud]

7 minutes ago, firedog said:

Unsuccessful is relative. Relative to some of the other rockers of their era 9.9 million is pocket change. Maybe the point is that wth the same music and a different business model a lot more money could have been made. 

 

Later in the thread I clarified that I was talking specifically about a band making a business success of its own recording label.  The vast majority don't, and that includes the Beatles and Dead.

 

More generally speaking, the Dead were forerunners of the current business model that uses recordings as advertising for the tours that make the money.

[Jud]

7 hours ago, lucretius said:

 

If by "*different*" you mean that it doesn't 'sound' like a derivative, then agreed; sometimes it takes a computer analysis to demonstrate it is a derivative.

 

I mean something we really would consider a different melody, with different chord progressions, but simply being in the same key.  On the other hand, something with the same chord progressions that we'd consider the identical melody, but in a different key, might well share fewer notes/pitches with the original.

 

I'm just pointing out that the way our brains match patterns can sometimes be rather unintuitive.

[Jud]

10 minutes ago, lucretius said:

 

Agreed. But why does this make it silly to seek a little more accuracy in music reproduction?

 

I'm sure we agree it isn't silly in the least.  But this goes back to the question @The Computer Audiophile asked about how measurements correspond to what we hear as "accurate," meaning most nearly reflecting what was recorded.  What I'm saying is we should be prepared for non-intuitive relationships there, in other words that the most common measurements (for example frequency response curves for speakers) may not be of the characteristics most important for our brains to think of something as "real."

[Jud]

23 minutes ago, lucretius said:

 

OK. I think the making it "real" part should be done at the mastering level, leaving consumer equipment manufacturers the task of creating equipment that will play the music as the mastering engineers intended (or as close as they can get).

 

Don't want to belabor the point, but I think a concrete example may help:

 

Some speakers, mine included, have a crossover design that is linear phase and is not dispersive, that is, time through the crossover filter is constant for all relevant frequencies.  This pretty much unavoidably creates a "hump" in the speaker's frequency response curve.  So which sounds more "accurate" to the brain: The localization and imaging allowed by the linear phase non-dispersive crossover design, or a speaker with a flatter frequency response curve that lacks these localization and imaging characteristics?  If you can't have both, which is the better goal for most realistic reproduction of the recording?

[Jud]

5 minutes ago, crenca said:

 

 

The transcendent third option - an active (DSP) crossover that get's you the linear phase with no FR hump.  Oh, our glorious have-our-cake-and-eat-it future  

 

Just about all the DSP I know of uses minimum phase rather than linear phase filters (also, I believe the Meridian apodizing filters were minimum phase, and I would be surprised if the MQA filtering *wasn't* minimum phase).

[Jud]

5 hours ago, crenca said:

 

So is this min phase filter being applied after the MQA processing magic, and if so does that not add back in phase errors which is a (the?) central SQ selling point of MQA?

 

4 hours ago, crenca said:

 

Yes, but if I am not mistaken a min phase filter adds phase errors.  So this filter takes away what MQA giveth...right?

 

No, it's part of MQA.  A minimum phase filter changes phase characteristics of the signal, but this is known and intentional.  If you called that a phase "error," the vast majority of speakers would be guilty of the same.

[Jud]

1 minute ago, mansr said:

Speakers are guilty of many things.

 

One wonders what they get up to when we're away.

[Jud]

17 minutes ago, Teresa said:

 

This deblurring has me really confused.

 

According to Wikipedia "Deblurring is the process of removing blurring artifacts from images, such as blur caused by defocus aberration or motion blur." 

 

So how does a digital or analog master audio recording become blurred? Perhaps they should figure out how masters become blurred and quit blurring them. Does this blurring only happen with heavily processed major label recordings? Or do even minimust naturally made recordings also suffer blurring?

 

Wouldn't it be better to not blur recordings in the first place?

 

And is this deblurring process available without MQA?

 

Hi Teresa.  Here's my attempt at a reasonable explanation:

 

The filters used in digital audio have a certain amount of what's called "ringing."  This is an ultrasonic signal that keeps going after the music signal ends.

 

Because it is ultrasonic, many people think it can't be heard.  However, some people believe it "smears" the underlying music signals, making transients less sharp.

 

Possibly in order to avoid this argument and thus an accusation they are claiming an effect people can't actually hear, MQA appears to be using the term "deblurring" rather than "minimizing ringing."  (This is supposing MQA filters at the mastering end of the process that no one outside of MQA knows anything about are actually minimizing ringing.)

[Jud]

Peter, your problem with shifting the ringing energy would be at least partly taken care of by an apodizing filter in the (re)mastering, as Miska points out.  I don't think it's being removed at the rendering end because everything we have seen about the filtering there is that it cuts too little.

 

Interesting point by Miska about slowing transients by removing the higher frequency components of the wave, though I'm sure many feel sensing that would be beyond human capabilities.

[Jud]

A couple of thoughts about unintended "benefits" of MQA, thinking about minimum phase filters:

 

- Such filters are to some degree dispersive, meaning how long it takes a part of the signal to pass through the filter varies by frequency.

 

- Whatever part of the ringing energy is not minimized by the apodizing aspect of the filter (if any) is moved to post-ringing, as Peter has pointed out.

 

Dispersive filters can contribute to a sense of depth; post-ringing might contribute to an overall sense of spaciousness (in a similar way as judiciously applied reverb does).

 

So how much of what people like about MQA might be due to an increased sense of depth and spaciousness?

[Jud]

29 minutes ago, Solstice380 said:

 

I think that the quantified effect of the ADC may be analogous to the lens error.  MQA does require a sample DAC to characterize so that they can assign / develop the appropriate filter for correction.  That's the way I read it.  

 

Yep, but you always want to be evaluating what's said versus what's known about filtering techniques and what's been disclosed by MQA in patents, white papers, etc. (as little as that may be).  And you also want to be careful about how far you stretch an analogy.  I think MQA has shown a degree of willingness to use some terms in marketing beyond their customary engineering meanings, e.g., "audibly lossless" to describe a lossy process where the usual term would be "audibly transparent" and "lossless" has a specific mathematical meaning that doesn't fit what MQA is doing.  (MQA is certainly not the only one and by no means even the worst at this among audio companies.)  It seems to me use of the term "deblurring" should fall into this category, when by all indications the usual engineering discussion would be about removing ringing by the well known technique of filtering out the ultrasonic frequencies where the ringing occurs.

 

There is also no particular magic in what MQA is doing at the (re)mastering or DAC end in terms of customization.  At the DAC end what appears to be happening filter-wise is very little  , presumably to avoid creating any ringing by the DAC.  This is at the cost of aliasing distortion, in an amount MQA judges to be insufficient to have much if any adverse audible effect.  At the (re)mastering end (though it must be admitted we can't know this for certain, as MQA has been fairly keen to restrict information about it), the only customization needed would be how much to roll off the ultrasonics depending on the amount of ringing produced by a prior ADC.  (Also, as @miguelito has pointed out, with the numbers of ADCs used in the production of most music these days, the idea of customizing to a particular ADC just doesn't work.)

[Jud]

10 minutes ago, Solstice380 said:

 

It is just a quantified offset, not trying to make it any more than it is.  We use calibration curves for almost everything, right?

 

There's no way to do such a specifically calibrated "offset" for time domain error caused by filtering.  (Peter mentioned this.) The way to remove ringing is by filtering out the frequencies at which it occurs, which is not the same thing as optical compensation and deblurring.  This is the danger in MQA's use of marketing terminology to describe engineering functions they feel are analogous.

[Jud]

10 minutes ago, Solstice380 said:

 

But MQA has said they are doing that, too.  

 

It may be the case for older music that has already been digitized.  But for newer music different ADCs are used in various parts of the recording and mastering process, so there's no way to compensate for the specific effects of a single ADC.  Besides, what is happening isn't counteraction and compensation in the time domain (not really possible), it's just filtering out the ultrasonic ringing.

[Jud]

53 minutes ago, jabbr said:

Wiener deconvolution: https://en.wikipedia.org/wiki/Wiener_deconvolution has long been applied to the audio domain (http://www.cnbc.cmu.edu/~samondjm/papers/Recio-Spinosoetal2005.pdf was found offhand as an example) and is widely used as a way to model and correct "error". In image analysis, a 3D model of the error is deconvolved with the 3D image to produce a corrected result. Similarly for 2d or 3d audio domain. The error may be measured from a system impulse response, or estimated. "deblurring" may refer to a sharpening in "linewidth" of fourier series peaks, but since MQA is proprietary who knows. The point I am making is that there are very well known and widely used techniques that are generally called "deblurring". In the audio domain, if one were producing, for example, a sonar map, then such techniques might be employed to sharpen such a map. This is analogous to "soundstage" (as an example).

 

Nothing in the MQA patents, white papers, or histories of the principals to indicate Wiener deconvolution or anything similar is going on, and lots in all of those to indicate it's good old Meridian apodizing filters in a new dress.

[Jud]

15 minutes ago, abrxx said:

 

I've certainly seen it said several times by Bob that he is NOT using Meridian apodizing filters. I believe even the relevant patents distinguish the method described therein fro the apodizing filters.

 

If you can point this out to me in the patents or white papers, I would appreciate it.

[Jud]

21 minutes ago, jabbr said:

 

I'm not surprised. Norbert Wiener's contributions have in many circles been ignored or have not been given the credit they are due. I think you would find his biography interesting. In many ways both Wiener (who at the time did not get the credit due) and Shannon both developed "information theory" independently and from different angles -- though both were at MIT. I am referring to the concept of a filter correcting error -- or eliminating noise see: https://en.wikipedia.org/wiki/Generalized_Wiener_filter -- given a modelled error (or noise). This is where the term "apodization" comes from. We used this technique: https://svi.nl/BeadsDeconvolutionExample in the 1980s. While Meridian may use a type of apodizing filter, they did not invent apodizing filters. I'd say that the *old* Meridian apodizing filters were, even at that time, Wiener filters in what at that time was a new dress

 

My suspicion is that the reason MQA remains "secret" is that the mathematical techniques are actually in very very wide use.

 

 

I doubt the Meridian filters use such sophisticated concepts.  They are relatively simple low pass filters designed to cut the frequency at which ringing may have occurred (the "corner frequency" of a previous filter) while not being so steep as to ring much themselves.  They are also usually minimum phase in order to put any ringing energy into post-ringing rather than equally into pre- and post-ringing as with a linear phase filter.

 

But unlike what I take to be one of a Wiener filter's functions, I don't think the Meridian filters are designed to remove noise from amongst signal components in the same frequency range.

[Jud]

2 hours ago, Fitzcaraldo215 said:

It seems to me that time blur is fundamentally about the distorted spread in time of the energy of an impulse caused by a-d and d-a in processing the digital signal.  According to the figure cited, that spread, the time blur, is much worse at 44k sampling than it is at higher sampling rates or with MQA.  I have no problem accepting that based on my own comparative listening.

 

See, this is what happens when the people marketing something don't use the usual terms for things.  This is a pretty exact description of ringing behavior.

[Jud]

28 minutes ago, esldude said:

All this speculation and no talk of subtractive dither?

 

Evidenced by its absence?

[Jud]

1 hour ago, pkane2001 said:

I'm familiar with deconvolution applied to images, but there is no obvious equivalency between optical distortions and audio distortions. What isn't clear to me is what is causing the blurring that requires deconvolving in audio. Convolution is a very specific mathematical function, and it must be shown first that any blurring function is truly mathematically convolved with the signal before you can attempt to apply deconvolution. 

 

I suppose you could work out the function of the Gibbs phenomenon for a given signal?  https://en.wikipedia.org/wiki/Gibbs_phenomenon  But why bother when you can remove the ringing that occurs as a result with a simple low pass filter?

[Jud]

Pictures of ringing caused by Gibbs phenomenon (downsampling 96KHz to 44.1KHz using three filters from an earlier version of Miska's Signalyst software):

 

 

 

 

[Jud]

Just now, esldude said:

Well subtractive dither is supposed to be part of MQA.  It is a genuine improvement in accuracy if done properly.  Whether audible or not it is better than additive dither.  Yet no talk of it.  I doubt things like the AQ DACs are able to handle their end of subtractive dither (but I don't know for sure).  Or is it always part of MQA? 

 

The best evidence about how audible the blurring is comes from Meridian.  The answer is not very.  Subtractive dither could be much bigger than this blur/deblur effect and completely non-controversial.  If it is really being done in MQA.  Certainly at least for new recordings it could be.

 

Known and used before MQA I assume?

[Jud]

Not from MQA:

 

Quote

We introduce quasi-subtractive dithering (QSD) as a practical way to approximate subtractive dithering (SD). Dithering is used in quantization to reduce correlation between the quantization error and the signal being quantized, usually at the cost of reduced signal-to-noise ratio (SNR). Theoretically, SD can achieve decorrelation without degrading SNR. Unfortunately, it requires the exact subtraction of the analog dither from the quantizer output (digital), which is impossible to implement in practice. An obvious compromise is to subtract from the quantizer output a quantized version of the dither. What is new in QSD is the introduction of a shrinkage factor that shrinks the quantized dither towards zero, which helps to control noise power.

 

 

[Jud]

5 minutes ago, mansr said:

I always strive for both, fun and profit. If nothing else, a good profit can enable various kinds of fun.

 

Perhaps not always?  Were you striving for profit when adding SDM capability to SoX?

[Jud]

4 minutes ago, mansr said:

That wasn't a business venture.

 

Not sure why I didn't think of that, but I didn't.  Fair enough.

[Jud]

11 minutes ago, The Computer Audiophile said:

 

Tidal is much more about fun and agenda than profits. 

 

Sure, but one wonders whether there might come a point of unsustainability.