MQA is Vaporware

[Jud]

5 minutes ago, Fitzcaraldo215 said:

As a former venture capital executive, I would not waste my time talking to someone who had an investment idea based on developing an open technology standard.  Sorry.

 

What proprietary technologies is Facebook based on?

[Jud]

Just now, Digital Assassin said:

"Among the “top questions” in the HBR article is the relationship between current and future open standards. In our opinion, when it comes to developing a successful digital business, open standards and open source create an ecosystem that benefits the core economics of a company. That article points out that, “While this may result in less control and less margins for the competitor, the reduced overall costs, the ability to scale, and the increased rate of innovation afforded by open source can outweigh the benefits of a proprietary platform.”

 

"...the increased rate of innovation afforded by open source can outweigh the benefits of a proprietary platform. A good example for this is how Google has created an open source platform with its Android operating system as compared to Apple’s more proprietary and closely controlled iOS."

 

https://hbr.org/2016/11/ceos-need-to-ask-the-right-questions-about-their-digital-businesses

 

 

 

 

 

Which shows that both can be successful.  Each has advantages and disadvantages.

[Jud]

Facebook is the peanut butter while the protocols are the jelly....

 

I think we can get a little too far into analogy.   The fact is that there is a continuum, from less to more open (e.g., Google and Apple both built OSs based on Unix; Google has been more open with its OS than has Apple; but Google is anything but open with its search algorithms).  

 

There is also the separate but related question of standards and interoperability, where MQA has been pretty terrible, and many successful companies have been quite good (MS and Apple frequently not having been good, but Google being a notable success that has usually been good at adhering to standards and maintaining interoperability).

 

It’s a little odd that folks who are worried about MQA taking over the world are at the same time arguing that those committed to standards and interoperability are usually more successful.  MQA, being less standards-compliant and interoperable, should therefore be less successful.  And in fact I think this is the case.

[Jud]

10 minutes ago, mansr said:

Yes, anyone can buy the CD spec from the IEC: https://webstore.iec.ch/publication/3885

 

Derivation of the term “RedBook.”

[Jud]

3 minutes ago, mansr said:

Moreover, if those alias artefacts are inaudible, what's the point in high-res, and by extension MQA, in the first place?

 

Depends on how audible one thinks time domain effects are, I suppose.  But that doesn’t necessitate MQA.

[Jud]

4 hours ago, Fokus said:

 

Correct, but only for the DAC side: these signals that trigger reconstruction filter ringing should not be present. But in reality they are, because the dominant type of AA filter used during music production is half-band, meaning that it is only a small multiple of 6dB down at Nyquist.

 

Moreover, the DAC side is not really the problem here. Look at the ADC side, where signals exceeding Nyquist are present prior to AA filtering. And thus these signals will make the AA filter ring.

 

For clarity: I believe, strongly (and based on experiments), that this ringing is not audible, provided it happens at a frequency not audible to the listener.

 

Would ultrasonic ringing be subject to the same potential problem as ultrasonic aliasing, intermodulation distortion?

 

Other than intermodulation distortion, is there anything you know of that would potentially cause time domain distortions in audible range transients?

[Jud]

Next questions, for anyone who knows:

 

1. Does the output of the MQA filtering in the DAC (a) go through the same upsampling to 8x rates and SDM as any other signal, or (b) bypass the normal upsampling and SDM?

 

2. If the answer to 1(a) is yes, does someone (Archimago or anyone else) have measurements showing any difference in analog output of a DAC vs. non-MQA?

[Jud]

8 minutes ago, Shadders said:

Hi,

I was of the understanding that an all pass filter had a constant time delay across the frequency band of interest, hence no smearing/delay/phase change in the frequency band of interest.

I would have thought that linear phase, which introduces a change in phase (albeit linearly) with regards to frequency, means, as an example, that at 100Hz there is no phase change (no delay), and at 10kHz there is a 45deg phase change, hence a delay when compared to 100Hz.

Is it that linear phase filters provide minimal temporal smear, but it still exists ?

Regards,

Shadders.

 

In linear phase filters, time through the filter does not vary by frequency.  In minimum phase filters, group delay is minimized but varies by frequency.

[Jud]

23 minutes ago, Shadders said:

Hi,

Thanks - but if the phase changes with frequency, and we have two signals in synchronisation, 100Hz and 10kHz, the 100Hz has no delay (0deg phase) but the 10kHz signal has 45deg phase (assume lag) and this introduces a 12.5uS delay ?

Regards,

Shadders.

 

I don’t know what the specific delays are, but yes, a minimum phase filter delays one frequency more than another, so it is a “dispersive” filter.  This is thought by some people to possibly give an illusion of depth.

[Jud]

32 minutes ago, mansr said:

Do you mean difference in an MQA DAC when playing normal content vs MQA? Of course it's different, mainly in the presence of high-frequency images and dither noise.

 

Interesting, I didn’t know whether some of the MQA ultrasonics might be removed in the process of upsampling to 8x (or higher) rates, SDM, and final reconstruction filtering.

 

If I remember correctly, you didn’t think the levels of ultrasonics you were seeing with MQA would be particularly audible.  Anything change your mind since?

[Jud]

2 minutes ago, Shadders said:

Are you sure this is correct.

 

Yep.

[Jud]

4 minutes ago, Shadders said:

but a linear phase filter incurs greater phase changes (assumed) albeit linearly with frequency.

Is a minimal phase filter called as such, since its phase change is less than other filters such as linear phase ?.

A phase change will equal time delay for that signal. They represent the same thing ?

Regards,

Shadders.

 

The *time delay* is constant over frequency with a linear phase filter.  Of course, as Fokus points out, equal time means different fractions of the wave at different frequencies.

[Jud]

8 minutes ago, Jud said:

 

The *time delay* is constant over frequency with a linear phase filter.  Of course, as Fokus points out, equal time means different fractions of the wave at different frequencies.

 

Note that this keeps the time and phase *relationships* between different frequencies constant.  These relationships are altered by minimum phase filters.  (How is the ultrasonic ringing “pushed” later in time than the main impulse?)

[Jud]

IBM is widely believed not to be doing terribly well with the open source and services model for a fair chunk of their business.

 

It’s very possible to succeed, fail, or anything in between with open source.

 

Where you will likely fail unless you achieve market dominance and can dictate is if your standards aren’t interoperable with others’.

[Jud]

1 hour ago, fung0 said:

MQA could take a long time to "fail," and do a great deal of damage along the way.

 

They could.  Currently, they seem to be taking a long time to succeed.

 

Vote with your dollars (as in, the absence of them from MQA).  It's a language that gets listened to.

[Jud]

1 minute ago, Fokus said:

Ironically, if MQA had been a bit more open from the beginning they would by now have been in a position with a much better chance for long-term success. But no ... they chose to be driven by maximum greed.

 

This is easy to say in hindsight.  But I think it was much more difficult to project at the time several years ago what would eventually allow for a return on investment.  In particular, I think they likely underestimated progress in streaming speeds, moved more slowly than they'd anticipated so streaming speeds had ramped up by the time they commercialized their ideas, or both, which negatively affected the value proposition of compression-while-retaining-nice-sound and left them to market on the basis of Fabulous Fidelity that really wasn't there.

[Jud]

2 hours ago, Shadders said:

Hi,

I have worked out my mistake, group delay is the differential of phase change across the frequency band. So constant group delay is linear phase change.

The time differences between a 100Hz and 20kHz signal for a 4th order bessel filter with cut off frequency of 22.05kHz is approximately 3.5uS - so not very much.

I have yet to simulate an 8th order elliptical filter - so i assume that this is what MQA is supposed to address - the temporal smearing of the signal due to the phase changes across the frequency band ?

Given the higher sample rates used today - i assume that temporal smearing is not an issue so we don't need MQA ?

Regards,

Shadders.

 

An experiment:  Set up a free test installation of Audirvana Plus.  Set upsampling preferences as you like, but leave the “pre-ringing” setting at 1.0 (linear phase).  Listen to whatever you like.  Then adjust pre-ringing to 0.0 (minimum phase) and see if you hear a difference.  (You may wish to upsample to the maximum input your DAC will accept, so as much of the upsampling as possible is performed by iZotope 64-bit SRC bundled with Audirvana Plus.)

[Jud]

57 minutes ago, Shadders said:

Hi,

Thanks. Seems to be a Mac only program. Also, do not have a DAC, but do have a surround sound processor - so testing is limited.

Regards,

Shadders.

 

You can also use a free test version of iZotope Rx itself and do the upsampling offline, then use mansr’s SoX fork if you wish to sigma-delta modulate to DSD format; or just use the SoX fork for the whole shebang.

 

You sure you don’t have a DAC, like reasonable sound card and headphones?  The point is to have not only something you can see, but something you can hear (or can’t).

[Jud]

3 hours ago, esldude said:

 

 

Funny is it not.  The solution to ringing that is at an inaudible frequency is all this rigmarole which creates audible band aliasing which we can cover with noise masking, and feel completely okay about as being higher fidelity when played back with leaky imaging filters that might also intermodulate into the audible band. 

 

At least if you manage to have a filter that leaks *and* rings, it can be quite audible.  With A+ I’ve managed to make myself oversampling filters that made music sound like someone had turned up the reverb way too high.  I haven’t gone back to do the experiment I recommended, otherwise using the default A+ settings, which should be non-leaky, and just play with phase (called pre-ringing in the A+ settings).  And I don’t have a free trial of iZotope Rx, though that route is open to others.  Or there’s the SoX fork.

 

Using A+ just makes it more convenient to adjust phase nearly “on the fly” with the very good iZotope resampling software, though it is (currently) Mac-only.

[Jud]

6 minutes ago, Shadders said:

Hi,

Thanks. Been out today, hence the late reply.

What i am not sure about is the pre-ringing - for a causal filter, surely there cannot be pre-ringing ? (or is it a non-causal filter) ??

Regards,

Shadders.

 

Doesn’t violate causality.  (It’s not reverb, so don’t think of it as that.)

[Jud]

46 minutes ago, Shadders said:

HI Jud,

I checked the books - and for the ringing to start before the main impulse, this is a non-causal system. As such, are you referring to digital filters (example) where non-causal filters can be implemented. (in general, not possible in the analogue domain).

Regards,

Shadders.

 

Hi Shadders.  Yes, the terminology may quite properly refer to "non-causal," but of course in a larger sense the laws of physics don't permit violations of causality.

[Jud]

By the way, Shadders, while looking for a good explanation for you of why "non-causal" filters don't violate the laws of physics (haven't found one yet as clear as I recall seeing a couple of years ago, so still looking), I ran across this interesting bit at the website of the Stanford University Center for Computer Research in Music and Acoustics (CCRMA), from the text "Introduction to Digital Filters with Audio Applications," by Julius O. Smith III, (September 2007 Edition):

 

Quote

Since we are listening to a lowpass-filtered impulse, it is reasonable to define the ideal expected sound as a "lowpass-filtered click," or some kind of "compact thump." We may therefore ask which signal sounds more like a lowpassed click? In the minimum-phase case, all filter ringing occurs after the main pulse, while in the zero-phase case, it is equally divided before and after the main pulse. Listening tests confirm that the "pre-ring" of the zero-phase case is audible before the main click, giving it a kind of "chirp" quality. Most listeners would say the minimum-phase case is a better "click." Since forward masking is stronger than backward masking in hearing perception, the optimal distribution of ringing is arguably a small amount before the main pulse (however much is inaudible due to backward masking, for example), with the rest occurring after the main pulse.

 

[Jud]

2 hours ago, Shadders said:

Hi Jud,

Thanks - do not have that book - but can you confirm (i think you did in the other post) that the pre-ringing is solely in the digital domain and is a non-causal filter ?

Else, are you indicating that non-causal filters exist in the analogue domain ?

Thanks and regards,

Shadders.

 

Sorry for the confusion Shadders, I'm not discussing filters at all, just fundamental physics that says the cause must precede the effect in time.  If you Google, you will see people who first run into the phenomenon of pre-ringing saying "How can this be - it violates the laws of physics to have an effect preceding a cause!"  And then people give more or less good explanations about how no, this isn't a violation of the very laws of physics on which the universe is based, this is how it works....

 

This is entirely aside from the fact that the specialized terminology that is used to talk about these filters indeed refers to them as "non-causal" due to pre-ringing.

 

The Smith book goes into this and more, and is available online.  Because it is from 2007 I don't know whether various bits are outdated, but you seem to be asking some fundamental questions that I believe may be answered there.  Here's a link:

 

https://ccrma.stanford.edu/~jos/filters/

[Jud]

2 hours ago, Shadders said:

Else, are you indicating that non-causal filters exist in the analogue domain ?

 

Below is a nice general explanation from a producer on the gearslutz forum.  The question he is answering is (summarized), "Why would anyone want to use linear phase filtering?  The pre-ringing messed up my kick drum sound."  Note he says nearly all analog filters are IIR.  But the crossovers in my speakers are linear phase filters, so they in fact are analog "non-causal" filters.

 

Quote

It's a good question, unfortunately the answer isn't very intuitive. But let's take a stab at it!

There are two general filter "architectures": 1. those with infinite impulse responses (IIR), and 2. those with finite impulse responses (FIR). IIR filters have infinite response because they use feedback -- the output depends on the input as well as previous outputs. Almost all analog filters are IIR. FIR filters, on the other hand, don't use feedback; the output depends only on the (scaled) input. FIR filters are trivial to implement digitally.

Now let's consider the phase response of a filter, which represents how the filter affects the phase of each frequency as the signal passes through it. One way to think about phase is as a frequency-dependent delay. Thus when we say that a filter is zero phase, we are saying that the filter passes each frequency without any delay (its phase response is a horizontal line at 0). When we say that a filter is linear phase, we're acknowledging that the filter will delay each frequency, but by a constant amount (phase response is a straight diagonal line over the passband). And if we say that a filter has nonlinear phase, we're saying that the delay is frequency-dependent (curved phase response).

It is a physical and mathematical property of filters that their phase response is determined by their impulse response symmetry. Picture the impulse response as an x-y graph, with time on the horizontal axis (x-axis) and amplitude on the vertical y-axis. If the impulse response (IR) is symmetrical about the point x=0, then the filter will be zero phase. If it is symmetrical about some other point other than x=0, then it will be linear phase. And if it does not have left-right symmetry, then the filter will be nonlinear phase.

Consider the zero-phase filter: it is symmetrical about x=0, which represents time at the "zero point". The zero point of time is of course the beginning of when we start our test (or measure our IR). But for the IR to be symmetrical, that must mean it looks the same at x=-1 as it does at x=1, and the same at x=-2 as it does at x=2, and so on. But negative numbers on the x-axis represent negative time, which (unfortunately!) we cannot access. So while it is possible to create zero-phase filters using clever tricks, it is much more straight forward to simply move the symmetry from x=0 to some positive number.

For example, let's say we have a 9-sample IR. If we center it at x=4, then the left-most sample in the IR will fall on x=0 and the right-most sample will fall on x=8. By shifting the IR to the right in time, we have maintained symmetry but lost the zero-phase property. Thus we have a linear phase filter. But remember that x=0 is time "zero", the beginning, so the shifting has introduced a delay of of 4 samples. In the real world, IRs are often several hundred (or even thousands) samples long, so the delays are correspondingly longer -- this is why you get latency when you switch to linear phase mode in your EQ.

Finally we can explain the "ringing". It's convenient (and accurate) to think of the IR as representing inertia in the filter. Like stretching a spring and letting go, the IR represents how the filter reacts to energy; it gives us a visualization in the time domain of how long it takes that energy to dissipate. This dissipating energy, in the form of oscillations, is what we call ringing, and all filters ring. IIR filters have nonlinear phase, which means they don't have left-right symmetry: the impulse response starts at its highest value with progressively damped oscillations trailing off to the right (in time), theoretically forever. Their ringing happens after the IR's "spike".

Linear phase filters, on the other hand, have left-right symmetry. If we compare the exact middle of the IR -- the "spike" -- with that of the non-symmetrical IIR filter, the left half of the linear phase filter can be seen as pre-ringing. This is usually easiest to hear on low-frequency transients (such as kick drum), where it sounds like the transient is being "smeared".

So why use linear phase filters at all? If we think of filters as physical systems responding to energy, it is easier to understand the tradeoffs of each design. The IIR filter is unruly in phase (and feedback can cause them to become wildly unstable), but its energy dissipation -- its ringing -- is easily masked because it always occurs after the impulse. FIR filters are perfectly stable and have beautiful phase response at the expense of dissipating their energy both in the forward and backward directions. It's a tradeoff, and the engineer must decide the appropriate tool for the task at hand. A linear phase filter is a bad tool on the kick drum, but it may be just the ticket on a top- and bottom-miked snare. Experiment and go with what your ear tells you, but a good rule of thumb is to use IIR unless phase response is critical.

Hope that helps; have fun!

 

[Jud]

Just now, Shadders said:

Hi Jud,

Thanks. I searched for the impulse response of analogue filters - they are all causal, and their impulse response is not symmetrical.

http://www.analog.com/media/en/training-seminars/design-handbooks/Basic-Linear-Design/Chapter8.pdf

The above text (gearslutz reference, not analog.com) i think is for digital filters - as the analysis mentions samples, and may be a non-causal implementation. Terminology and context seemto be quite important here

Regards,

Shadders.

 

OK, I could easily be wrong.  I do know they are linear phase.