John Atkinson: Yes, MQA IS Elegant...

[Shadders]

23 hours ago, John_Atkinson said:

So as very few commercial recordings have been made with the Ayre QA-9's Listen filter, it would appear that if removing so-called "temporal bur" is indeed something that improves sound quality for the reasons I explain in the article,  the MQA process is one of the few commercially available end-to end solutions that would do that. If...

Hi,

What this implies is that, to remove the "temporal blur" (dispersion is the correct engineering term for this) you need an end to end solution.

 

Therefore, ALL processing by MQA on existing recordings is pointless and may mess up the recording. That is, MQA can NEVER be backwards compatible with existing recordings.

 

Regards,

Shadders.

[Shadders]

1 hour ago, mansr said:

With or without spelling mistakes, that concept still doesn't make sense to me in the context of sampling and reconstruction.

Hi,

Is this not referring to matched filters - root raised cosine at each modem which when multiplied together is a raised cosine filter - so this is an analogy to the filters in the audio chain, at recording, and playback ?

Regards,

Shadders.

[Shadders]

49 minutes ago, mansr said:

I'm familiar with matched filters in other contexts. I just don't see how the concept applies here.

Hi,

An initial thought is that the filter has a less impact on the recording, and the playback, but once the two are used together in the chain, the result is the combination - which is heard. The combined filter has less impact, than the playback filter being of any design of another manufacturers DAC - which could be any design.

 

All it is, is Ayre presenting the proposal as per MQA, but it is free. Does not have to be Ayre's filter - could be any filter in the recording, but if it is known, then the DAC can be designed appropriately.

 

I do not see the DAC IC (ADC/DAC semiconductor) manufacturers responding to MQA or taking any interest in the audio field - apart from the smaller, more bespoke/esoteric manufacturers.

 

Regards,

Shadders.

[Shadders]

27 minutes ago, mansr said:

Which aspects of the filters should be matched, and how, in order to achieve this?

 

The thing is, reconstruction is not really the inverse of sampling. The perfect anti-aliasing filter used during sampling is an ideal low-pass filter. The perfect anti-imaging filter during reconstruction is also an ideal low-pass filter. Obviously, ideal filters are impossible to realise. However, nothing says they need to be matched in any way. If both have a stopband attenuation of at least 200 dB and a passband ripple of at most 0.01 dB, which is easily done with digital filters, there is nothing to worry about.

 

Any matching of the filters would have to be related to their deviations from ideal low-pass filters, so let us consider the effects of imperfections. Lacking stopband attenuation in the anti-aliasing filter will result in high frequencies aliasing irreparably into the audible range. No amount of "matching" in the reconstruction filter can make up for this. Similarly, poor stopband attenuation in the anti-imaging filter creates images of the low frequencies regardless what the anti-aliasing filter looked like. Unlike aliasing, however, images fall outside the audible range and are thus mostly harmless unless they are strong enough to cause audible intermodulation distortion or put excessive power demands on amps and tweeters.

 

In the passband, final output signal contains the combined imperfections (ripple and phase shifts) of the two filters. If the anti-aliasing filter is so poor as to have audible anomalies here, a reconstruction filter carefully chosen to have precisely the opposite ripple, a peak in one wherever the other dips and conversely. A much more practical approach, however, would be to simply correct any such effects before distributing the digital streams rather than trying to get a reconstruction filter with matched horridness into every DAC. Or use a decent ADC in the first place.

 

Now I want to emphasise that the anti-aliasing filters commonly used do not have any passband anomalies that need correcting. It is a non-problem. Moreover, there is no indication, from MQA or Ayre, that this is the kind of matching they're talking about. Indeed, their filter designs, near as I can tell, do quite the opposite, the reconstruction filters exacerbating early roll-off and phase shifts already present from the recording filters.

Hi,

It was an initial thought by me as to what Ayre are referring to.

 

The current ADC's from what has been stated on this forum (by others), sample at a much higher rate than 44.1kHz. Therefore the ideal filter is not required to meet the restricted bandwidth to ensure no aliasing. The filter for ADC can be relaxed - however that is to be designed and implemented. The ideal filter is not required as we are not using a low sample rate.

 

Given this, my interpretation is that Ayre are proposing an end to end system which has minimal impact on the music, but their system is free, in that is not patented or protected by IP.

 

DAC IC manufacturers can follow the proposal, or not.

 

Let us wait and see how @Ryan Berry responds to what he was actually referring to.

 

Regards,

Shadders.

[Shadders]

Off topic - deleted.

 

[Shadders]

15 minutes ago, crenca said:

Interesting, perhaps Charley did mean something other than high(er) sample rates by "complimentary".

 

[Shadders]

56 minutes ago, Le Concombre Masqué said:

You're right : there are more fundamental things to do when you have the privilege to talk to music producers, recording and mastering engineers as MQA people claim they do, such as making sure they use a calibrated  Stereo system and calibrating yours. However, I think loudspeakers can perform better than what you expect ; try REW + Rephase ; I get this from my Cabasse (30 years old dynamic drivers, pre SCS) 

Hi,

Yes - speaker distortion can be single digit figures, so my interpretation is that any changes to mastering, or the use of high resolution(PCM, DSD, MQA) is completely lost.

As you have stated, REW etc., has such a major change in sound, that any gains from CD to high resolution are minimal by comparison. I have seen many reports that using miniDSP had such a positive benefit to the sound.

 

In the big scheme of things, the continued support and promotion of MQA is comical. No one in the normal world cares about high resolution - let alone a new format. People are happy with MP3. The revenue from extra sales of MQA must be non existent.

 

So why do some Hifi spokes people persist in promoting MQA ?.

 

Like VHS to DVD, the step change was major and everyone could see it, so too was the change from LP to CD (hear it).

 

Why the record companies are even supporting MQA has to mean that there is another reason other than sound quality. They must be aware of the decimation of the hifi industry, where high resolution is becoming inconsequential.

 

So, to @John_Atkinson why do you still promote MQA in any fashion, if no one really cares (in the real world) ???.

 

[i assume that it is because your job is in hifi, and it is just another wheeze].

 

Regards,

Shadders.

[Shadders]

44 minutes ago, tmtomh said:

There does seem to be one major logical (or perhaps evidentiary) gap in the piece, though: He cites a listening test that failed to turn up audible evidence that pre-ringing causes problems (except in an intentionally worst-case filter where all the ringing is pre-ringing). Then almost immediately he nevertheless speculates that pre-ringing nevertheless could be an issue (by wondering if higher sample rates are pleasing to folks because they move ringing out of the audible range) - and then quickly moves on from there to strongly imply that pre-ringing is indeed a problem, despite the lack of evidence in the listening test mentioned just a couple of paragraphs above.

Hi,

From:

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

 

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.

 

Regards,

Shadders.

[Shadders]

3 minutes ago, mansr said:

That's testing a filter with a cutoff frequency of 2 kHz. Of course the ringing is audible, especially when listening to the worst case, the impulse response itself. Move it up above 20 kHz, and things change.

Hi,

One aspect is, that the issue of ringing when audible has been tested and results provided. Even at 20kHz cut off, ringing exists. Maybe you cannot hear that frequency ?

 

Have you analysed the ringing ?. If you use Octave and implement a linear phase filter for example, and a very slow ramp, 0volts to peak 1volt in 1 second, there is still ringing - albeit extremely low level - you have to zoom into the waveform. I have not analysed how this transfers into 16bit word, or 24bit words.

 

The ringing amplitude is directly proportional to the rate of change of the input signal. It has an envelope too, and is not just a pure cut off frequency tone.

 

Regards,

Shadders.

[Shadders]

Just now, mansr said:

Of course I can't hear 20 kHz. Nobody above the age of 12 can.

Hi,

People can at the age of up to 25 can hear at 17.4kHz - there are security speakers - the Mosquito anti-loitering :

 

https://en.wikipedia.org/wiki/The_Mosquito

 

I expect that there are some in this age range from 18 to 25 who can hear 20kHz

 

Regards,

Shadders.

[Shadders]

8 hours ago, mansr said:

I haven't been 25 in a long time. Besides, not even an 18-year-old can hear 20 kHz sounds 60 dB or whatever below the music in the sub-10 kHz range. That page you linked proves nothing of relevance whatsoever.

Hi,

You have no proof that a person in the age range 18 to 25 cannot hear 20kHz. The mosquito product works at 17.4kHz so as to ensure it works on most humans up to the age of 25.

 

I have no proof that every 18 to 25 year old can hear 20kHz, but there will be some people in that range that can.

 

The original point of my post is that ringing preferences have been characterised, regardless whether people can hear 20kHz or not.

 

Regards,

Shadders.

[Shadders]

17 minutes ago, mansr said:

Nobody is denying that young people can hear up to 20 kHz. However, even in a child, their sensitivity at 20 kHz is nowhere near what it is below 10 kHz. Now look at the frequency spectrum of actual music. In a typical recording, the level at 20 kHz is 30 dB or more below the intensity of any frequency in the main content region. If the total power of the actual music is compared to that of the "ringing," the difference is much larger. A test at 2 kHz tells absolutely nothing about the audibility of these effects at 20 kHz. Besides, for CD audio, we're talking about 22 kHz, and very few people indeed can hear that high, even when young. Still, let's say there is a slim chance that a 4-year-old with bat-like hearing might find the "ringing" displeasing. To avoid this, simply move it up to 48 kHz. There nobody can possibly hear it.

Hi,

Again, the point i was referring to is that ringing has been characterised - and the preference as given in the website.

 

If you are stating that the ringing at the higher frequencies do not correlate to the same preference at 2kHz, then ok.

 

Regards,

Shadders.

[Shadders]

21 minutes ago, mansr said:

Given what we know about human auditory perception, it does not appear valid to generalise the findings for 2 kHz so as to apply at the very extreme of our hearing ability.

Hi,

So no evidence either way that it is, or is not, applicable.

Regards,

Shadders.

[Shadders]

16 minutes ago, crenca said:

 

The only evidence I know of about this audibility of this very low level, high frequency "ringing" is provided by audiophiles - and even they don't speak of terms of being able to directly hear it - rather they talk about how they can "hear" how transients are "smeared" and the like, and then they go on to reason a correlation between this "smearing" and impulse tests and the "ringing" these tests reveal.  

 

In the other words, it's all one big audiophile house of cards that in theory could be true (how would you test it outside of subjectivsed audiophiledom and its reliance on sighted listening "tests"?) but probably is not...

Hi,

OK, if you examine the ringing - it is not just a single frequency at a constant amplitude. The ringing has an envelope - this envelope has a frequency content predominantly less than the ringing frequency. It has to, else it would not be an envelope.

 

It is possible that this is what "audiophiles" are hearing, the effect of the ringing envelope.

 

Regards,

Shadders.

[Shadders]

7 minutes ago, crenca said:

 

Ah, intriguing question. The envelope itself is still very low level in amplitude is it not (honest question)?  In the video I posted above:

 

https://www.computeraudiophile.com/forums/topic/49609-john-atkinson-yes-mqa-is-elegant/?page=5&tab=comments#comment-864679

 

They force (through extreme Q EQ, gain, etc.) the ringing of a linear (0 phase) filter to be audible, and while it is a chirping sound it certainly is below 8000khz I believe.  Is this the envelope and and not the ringing frequencies themselves that have become audible?  

 

Hi,

I used Octave to analyse the envelope - frequencies are across the frequency range up to the ringing frequency.

 

I said earlier that ringing occurs in every waveform that has a transient change - the transient does NOT have to be at the ringing frequency - i said it was based on the rate of change of the signal, I should correct this - it is a second order effect - it is the rate of change of the rate of change of the signal, that causes ringing. The transient response.

 

If the music content contains transients then ringing will occur - it is difficult to see once the waveform is non-zero - it is embedded in the music signal - low level.

 

If you have a play with Octave, or Matlab - you can see the effect. Whether this is the smearing that they are talking about - who knows.

 

Regards,

Shadders.

[Shadders]

30 minutes ago, crenca said:

 

Just to be clear Shadders, when you say you can "see the effect" you mean with an impulse or impulse like "transient" that comes out of 0dBFS?  You don't mean you can see the effect in the middle of an otherwise normal musical signal/waveform?

 

 

Hi,

I initially used a sine wave at 1kHz which was gated - started from zero and finished at zero, and saw the pre and post ringing. I analysed the ringing from that to see the envelope. I examined the initial ramp of the sine - the ringing continued to appear (as it should) embedded here too. Essentially the amplitude of the sine signal was modulated for 256 samples assuming a 512 tap filter, reducing in level of modulation as the sine waveform progressed.

 

I then examined a ramp signal from 0volts to 1volts, which took 1 second to reach 1 volt, using a sample rate of 192kHz, and examined the pre-ringing. It appears even in this signal at an extremely low level.

 

This is obvious - if you look at the coefficients of the filter - you can see that it will present the +ve/-ve values accordingly.

 

The filter will have a steady state response and transient response. The steady state response is just that - when a signal is passing through such as a sine wave - steady state (after N+1 samples for an N tap filter) - the output will have an attenuation and phase characteristic. Same for a step change - the steady state response is the final settled output which will be the step voltage.

 

For the transient response - the ringing will depend on the amplitude and rate of change of the transient. The ringing is small - in many cases, and music by its very nature has many transients, negative and positive - where they will oppose or reinforce . You will not see obvious examples of ringing through a continuous (unbroken waveform) of music - since it is embedded, and small in amplitude.

 

Use Octave/Matlab - to examine the ringing in waveforms - as the waveform starts to build from zero, ringing is still in the waveform - embedded - but very low level.

 

Regards,

Shadders.

[Shadders]

37 minutes ago, mansr said:

This is not a band limited signal.

Hi,

Yes, i know. It demonstrates the ringing to analyse the envelope.

There will be transients in music from minimal to a maximum - and the envelope amplitude will be proportional to the transient.

 

If you examine a 1volt peak sine wave at 1kHz which reaches 0.5volts and then jumps to 1volt in the next sample, it will be a transient and you will see ringing.

 

In the next test, the 1volt peak is reached in 2 samples (0.5v, 0.75v, 1.0v). You will see ringing. Then the 1volt peak is reached in the next 4 samples (0.5v, 0.625v, 0.75v, 0.875v, 1.0v) you still see the ringing - reduced in amplitude.

 

If you keep doing this, you will eventually have a tangent to the sine curve from 0.5volts - no ringing at the 0.5volts sample.

 

So, from the worst case which is next sample at 1.0volts, to the tangent, the ringing occurs, but reduces to zero at the tangent test. This shows that ringing is inherent in the waveform, and appears at every transient, where that transient does not need to be band limited.

 

Regards,

Shadders.

[Shadders]

1 minute ago, mansr said:

I don't understand what you're trying to prove. Nobody is saying that filters don't "ring" when applied to impossible signals.

Hi,

If this is in regards to the gated sine, i agree - was not band limited.

 

The following text in the post demonstrates that it does not need to be an out of band signal to cause ringing.

 

Ringing is inherent in the filter - see my previous post on transient response and steady state response.

 

Music is full of transients, and they do not need to be out of band to cause ringing. That ringing is embedded in the signal - difficult to see, and does not look like the pure ringing as seen by an out of band signal response.

 

Regards,

Shadders.

[Shadders]

1 minute ago, mansr said:

Not in the mathematical sense.

Hi,

I am not sure what you are trying to prove here.

https://en.wikipedia.org/wiki/Transient_(oscillation)

There does not seem to be a stated bound for a transient.

 

I will assume you mean that a transient has to represent energy outside the bandwidth of the the system.

 

Can you provide the definition of a transient ? Thanks.

 

Anyway, if you disagree with the presented example, description, or whatever, then ok  - i am happy to read your thoughts.

 

Regards,

Shadders.

[Shadders]

56 minutes ago, Fokus said:

Take a  'ringing' Sinc-like low-pass filter with a transition frequency F.

 

If the spectrum of the stimulus is non-zero at F, then the output signal will visually exhibit filter ringing. If the spectrum of the stimulus is zero at F, then the output signal will not show ringing.

 

It is as simple as that

Hi,

This seems to be a mantra on this site. People seem to be stuck on boundary conditions/effects.

 

If you examine my example of the sine wave with a discontinuity when it is 0.5volts going to 1.0volts at the next sample, and you relax the discontinuity until it is a tangent to the curve, ringing is always there, but it reduces in amplitude.

 

Examine my ramp example - easy to implement in Octave/Matlab - ringing is always there no matter how small a change, but the ringing is extremely small.

 

Examine the filter action in the time domain, look at the coefficients, and recall the convolution theorem where each input sample to the filter produces a scaled impulse response, which are summed together in time, each have the relevant delay. This should give you some insight to what is actually occurring.

 

Examine the DSP books and look up the aspects of transient and steady state responses. Every DSP book has this discussion and analysis. Every electrical engineering book has this discussion when dealing with filters, or any other circuit where a transient is applied - usually studied using Laplace transforms, most people just use spice now when analysing.

 

OK - in saying this - i know this will be ignored or scoffed at, and the mantra will win through.

 

1 hour ago, Fokus said:

As for the ringing envelope occupying frequencies below F ... consider that a filter is a linear operator, and think this through.

Not sure what you are saying here. A linear system will pass the envelope if the frequencies are in the passband as per the filter transfer function.

 

Regards,

Shadders.

[Shadders]

3 minutes ago, Fokus said:

 

That is abundantly clear.

 

Once more ... you claim that you investigate the ringing in isolation, and you claim that the ringing envelope emcompasses frequencies down in the passband. In other words, you claim that the ringing itself has audible content all over the passband. This means that the filter creates new spectral content, well below the filter's transition frequency.

 

Now tell us the main properties of a linear operator.

 

Hi,

Are you are referring to the filter being a linear system that adheres to the superposition theorem ? That is :

 

Assume T is the operator :

 

T[a1.x1(n) + a2.x2(n)]   = a1.T[x1(n)]    +   a2.T[x2(n)]

 

Or are you stating that the response of the filter is linear ?

 

Regards,

Shadders.

[Shadders]

5 minutes ago, Fokus said:

A linear operator cannot create new spectral content.

Hi,

Ok. When you state linear operator, do you mean a linear system ?

Regards,

Shadders.

[Shadders]

1 minute ago, Fokus said:

A linear, time invariant system, ... like a filter.

Hi,

OK - then this applies :

A linear system that adheres to the superposition theorem. That is :

 

Assume T is the operator :

 

T[a1.x1(n) + a2.x2(n)]   = a1.T[x1(n)]    +   a2.T[x2(n)]

 

The other criteria i did not mention is that the system has to be relaxed. Then given this, i would agree, that the output of the system can only be linearly related to the input.

 

If you then examine my examples of the ramp, or the sine with discontinuity which is relaxed to a tangent, where both exhibit "ringing", then there is something else occurring.

 

Maybe it is the term "ringing" that is the problem, where this is always attributed to the impulse response, or out of band energy applied to the filter.

 

In any case, the distortion occurs with transients.

 

Separate issue, the envelope of ringing does have energy within the passband.

 

Regards,

Shadders.

[Shadders]

17 minutes ago, Fokus said:

I was not asking you what a linear system is.

 

I asked the question for you to ponder what this means in the face of your claim

 

" Separate issue, the envelope of ringing does have energy within the passband. "

 

Hi,

I am in agreement with you - that ringing if caused by an impulse, is purely the filter response, and when transformed to the frequency domain will be the frequency response of the filter. As such, the ringing does indeed have frequencies from 0Hz to the transition band - based on its waveform which includes its envelope.

 

Regards,

Shadders.

[Shadders]

1 minute ago, mansr said:

That's just a very roundabout way of stating that if the input contains all frequencies, as is the case with an impulse, the output will also contain all frequencies up to the filter cut-off. If it didn't, you'd have a high-pass filter.

Hi,

Maybe, just needed to state where there was an agreement.

Regards,

Shadders.