Jump to content

What MQA might be...


Recommended Posts

It is possible that MQA is based on an end to end deconvolution. I've posted this blog

 

The technique of room correction can be applied to the entire digital signal chain. That would be "temporal deblurring".

 

If we were to create such a system, we would end up with similar minimum equipment specs as MQA requires.

Custom room treatments for headphone users.

Link to post
Share on other sites
It is possible that MQA is based on an end to end deconvolution. I've posted this blog

 

The technique of room correction can be applied to the entire digital signal chain. That would be "temporal deblurring".

 

If we were to create such a system, we would end up with similar minimum equipment specs as MQA requires.

 

You cannot "unfilter" the ADC, so at most you can use apodizing filters for deblurring as has been used for long time already. Same goes for the DAC so you cannot unfilter the analog reconstruction filter that is after your process.

 

What you can do is to correct for engineering mistakes of those if they have bad frequency- or phase response. With oversampled ADCs and DACs the effects of those in well engineered gear is minimal though.

 

However, this doesn't require a proprietary DRM'd delivery mechanism. The ADC correction can be performed at source and then encoded as normal FLAC. The DAC correction can be performed at playback time, in fact I could add such profiles to HQPlayer for all the DACs I have.

 

There is no need to tie the two together or have anything "end-to-end".

 

When it comes to end-to-end solution without blurring problems of decimation/oversampling digital filters and without DRM, it has been around for years and it is called DSD...

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to post
Share on other sites

However, this doesn't require a proprietary DRM'd delivery mechanism. The ADC correction can be performed at source and then encoded as normal FLAC. The DAC correction can be performed at playback time, in fact I could add such profiles to HQPlayer for all the DACs I have.

 

Exactly.

 

When it comes to end-to-end solution without blurring problems of decimation/oversampling digital filters and without DRM, it has been around for years and it is called DSD...

 

;)

 

There probably is a role for kernels for different earbuds, headphones, amps, speakers etc. Assuming sufficient headroom and bandwith, a lot could be done, for example, make one type of amp sound like another, expand the soundstage etc., model capacitors, coils etc.

 

There seems to be good recognition and excitement about the abilities of deconvolution to do room correction (spatial domain), similar abilities in time domain and more.

Custom room treatments for headphone users.

Link to post
Share on other sites

An example where an analogous technology is used in digital photography would be DXO Optics Pro -- they provide profiles for a wide variety of lenses and camera sensors. There are also filters that can be applied to the image and finally output profiles are applied to the image for printing on specific printers and using specific papers. Each of these three phases are distinct and can use different specialized software packages.

Custom room treatments for headphone users.

Link to post
Share on other sites
An example where an analogous technology is used in digital photography would be DXO Optics Pro -- they provide profiles for a wide variety of lenses and camera sensors. There are also filters that can be applied to the image and finally output profiles are applied to the image for printing on specific printers and using specific papers. Each of these three phases are distinct and can use different specialized software packages.

 

Nikon Capture NX has also lens profiles and I believe some others too. My Pentax DSLR also has built-in profiles for their own lenses.

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to post
Share on other sites
There probably is a role for kernels for different earbuds, headphones, amps, speakers etc. Assuming sufficient headroom and bandwith, a lot could be done, for example, make one type of amp sound like another, expand the soundstage etc., model capacitors, coils etc.

 

There seems to be good recognition and excitement about the abilities of deconvolution to do room correction (spatial domain), similar abilities in time domain and more.

 

There the difference and possibilities are much bigger and more important than at ADC or DAC, because the changes are like order of magnitude higher. One can use for example Acourate or Audiolense to create a filter for the entire playback chain and then apply it while playing. I can almost guarantee that there are no problems hearing the difference in such case!

 

For just profiling the DAC, one can run Acourate through simple analog loop-back as long as the ADC is calibrated first. Or one can also loop back from the power-amp with suitable gear if preferred.

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to post
Share on other sites
An example where an analogous technology is used in digital photography would be DXO Optics Pro -- they provide profiles for a wide variety of lenses and camera sensors. There are also filters that can be applied to the image and finally output profiles are applied to the image for printing on specific printers and using specific papers. Each of these three phases are distinct and can use different specialized software packages.

 

Probably a pretty good analogy as even with the very best equipment (I use the Canon 1D series) the filters available in software contribute to a much better final photograph, although as with music, they can both be used to remove true camera artifacts and to "master" the file into something less accurate, but perhaps more pleasing to the eye.

Synology NAS>i7-6700/32GB/NVIDIA QUADRO P4000 Win10>Qobuz+Tidal>Roon>HQPlayer>DSD512> Fiber Switch>Ultrarendu (NAA)>SMSL M500 DAC> Bryston SP3 pre>Levinson No. 432 amps>Magnepan (MG20.1x2, CCR and MMC2x6)

Link to post
Share on other sites
You cannot "unfilter" the ADC, so at most you can use apodizing filters for deblurring as has been used for long time already. Same goes for the DAC so you cannot unfilter the analog reconstruction filter that is after your process.

 

What you can do is to correct for engineering mistakes of those if they have bad frequency- or phase response. With oversampled ADCs and DACs the effects of those in well engineered gear is minimal though.

 

 

You touch upon a point that I have been curious about so I want to seek more clarification as I am not totally on top of what can be accomplished or corrected through DSP. I understand apodizing filters changes the occurrence of the ringing that a steep digital filter introduces from happening before an impulse to happening after an impluse and hence has the potential to sound more natural. But if that is correct, the blurring remains but is less offensive. Can the apodizing filters be tuned somehow to remove the ringing if the filter in the ADC is known?

When you say you can not unfilter the ADC, do you mean that any negative effects introduced by the filters in the ADC, including ringing/smear, can not be corrected except for frequency and phase response?

Link to post
Share on other sites
You touch upon a point that I have been curious about so I want to seek more clarification as I am not totally on top of what can be accomplished or corrected through DSP. I understand apodizing filters changes the occurrence of the ringing that a steep digital filter introduces from happening before an impulse to happening after an impluse and hence has the potential to sound more natural.

 

Apodizing filters just replace original ringing with something else. In many cases, like Meridian's, it is minimum phase so it is all post-ringing and thus masked by natural decay. But it can also be linear phase or something else. I have both types of apodizing filters I choose based on source content genre.

 

But if that is correct, the blurring remains but is less offensive. Can the apodizing filters be tuned somehow to remove the ringing if the filter in the ADC is known?

 

Not all ringing, because as long as you have steep filter, there is also ringing. Due to very slow roll-off of, slow roll-off order filters are really useful only at very high sampling rates, such as DSD.

 

When you say you can not unfilter the ADC, do you mean that any negative effects introduced by the filters in the ADC, including ringing/smear, can not be corrected except for frequency and phase response?

 

To some extent yes. We are now talking about two category of filters:

1) Decimation filters

2) Interpolation filters

 

In PCM system, (1) removes any information that doesn't fit into the Nyquist band, that is half of the sampling rate. These are used in oversampled ADCs, or if you for example convert DSD to PCM. If there is nothing to remove vs input signal, there is no ringing either and no loss of information (given theoretically perfect implementation). In practice, CD format ends up removing source signal parts almost always, while 192k hires rarely removes anything and DXD (352.8k) practically never. Given that DSD64 has Nyquist bandwidth of 1.4 MHz, there are not practical needs to particularly limit the input bandwidth. What ever has been removed is lost forever, the information is just gone. You can reduce the ringing afterwards, and make it less audible, but the removed information is not going to come back, so you cannot remove effect of the filter in that sense.

 

(2) is used at DAC side and doesn't remove any information, it just removes duplicates of the existing information that repeat at every multiple of the sampling rate (inherent property of PCM sampling). So a CD repeats it's information around every multiple of 44.1 kHz sampling rate. This stage can also modify impulse response of (1) using an apodizing filter. The sampling rate is like a mirror. Below the sampling rate there's a reverse mirror image of the content and above sampling rate there is normal mirror image of the content. When interpolated, the images become less frequent. So for example if the content is properly interpolation filtered to 352.8k sampling rate, the mirror images repeat only around every multiple of 352.8k. Analog reconstruction filter after the conversion removes then those images. It is easier to make an analog filter that completely removes an image that first appears at 352.8k and then at every multiple than it is to make an analog filter that would remove an image that first appears at 44.1k and then every multiple of that. Removing effect of this reconstruction filtering would also remove the actual reconstruction phase...

 

 

DSD and other non-Nyquist heavily oversampled noise-shaped sampling systems operate in somewhat different way.

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to post
Share on other sites
Apodizing filters just replace original ringing with something else. In many cases, like Meridian's, it is minimum phase so it is all post-ringing and thus masked by natural decay. But it can also be linear phase or something else. I have both types of apodizing filters I choose based on source content genre.

 

 

 

Not all ringing, because as long as you have steep filter, there is also ringing. Due to very slow roll-off of, slow roll-off order filters are really useful only at very high sampling rates, such as DSD.

 

 

 

To some extent yes. We are now talking about two category of filters:

1) Decimation filters

2) Interpolation filters

 

In PCM system, (1) removes any information that doesn't fit into the Nyquist band, that is half of the sampling rate. These are used in oversampled ADCs, or if you for example convert DSD to PCM. If there is nothing to remove vs input signal, there is no ringing either and no loss of information (given theoretically perfect implementation). In practice, CD format ends up removing source signal parts almost always, while 192k hires rarely removes anything and DXD (352.8k) practically never. Given that DSD64 has Nyquist bandwidth of 1.4 MHz, there are not practical needs to particularly limit the input bandwidth. What ever has been removed is lost forever, the information is just gone. You can reduce the ringing afterwards, and make it less audible, but the removed information is not going to come back, so you cannot remove effect of the filter in that sense.

 

(2) is used at DAC side and doesn't remove any information, it just removes duplicates of the existing information that repeat at every multiple of the sampling rate (inherent property of PCM sampling). So a CD repeats it's information around every multiple of 44.1 kHz sampling rate. This stage can also modify impulse response of (1) using an apodizing filter. The sampling rate is like a mirror. Below the sampling rate there's a reverse mirror image of the content and above sampling rate there is normal mirror image of the content. When interpolated, the images become less frequent. So for example if the content is properly interpolation filtered to 352.8k sampling rate, the mirror images repeat only around every multiple of 352.8k. Analog reconstruction filter after the conversion removes then those images. It is easier to make an analog filter that completely removes an image that first appears at 352.8k and then at every multiple than it is to make an analog filter that would remove an image that first appears at 44.1k and then every multiple of that. Removing effect of this reconstruction filtering would also remove the actual reconstruction phase...

 

 

DSD and other non-Nyquist heavily oversampled noise-shaped sampling systems operate in somewhat different way.

 

And the above is true because actual filter ringing only occurs in the transition band, is that correct? If your transition band for 192 khz sampling is between 80 khz and 96 khz, rarely is there anything actually there to ring the filter. With DXD the transition band is 160 khz to 176 khz which is near certain to have nothing from an audio recording which might ring the filter.

And always keep in mind: Cognitive biases, like seeing optical illusions are a sign of a normally functioning brain. We all have them, it’s nothing to be ashamed about, but it is something that affects our objective evaluation of reality. 

Link to post
Share on other sites
And the above is true because actual filter ringing only occurs in the transition band, is that correct? If your transition band for 192 khz sampling is between 80 khz and 96 khz, rarely is there anything actually there to ring the filter. With DXD the transition band is 160 khz to 176 khz which is near certain to have nothing from an audio recording which might ring the filter.

 

Yes, exactly...

 

Why something A and B could still sound different at DXD is due to other factors, like differences in the sigma-delta modulators, analog stage differences, intermodulation from images around multiples of 352.8k sampling rate if not oversampled properly before the modulator, etc. But almost certain not the digital filter ringing... And MQA does not address the modulator or oversampling above 352.8k, etc.

 

My current thinking is that sound of a DAC pretty much depends 33% on digital filters, 33% on modulator and 33% on the conversion/analog stage...

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to post
Share on other sites

Is this correct? "actual filter ringing only occurs in the transition band"?

 

The transition band in the usual 20KHz brickwall filter is typically from 20Khz to 22KHz, right? Or more generally, it's the frequency band between the passband & the stopband

 

With steep filters (usually halfband equiripple filters) isn't there ripple in the passband i.e before 20KHz?

 

Isn't this where the concept of impulse response time-smearing come arising from ?

 

If the ideal is that the passband should be perfectly flat, can the ADC passband ripple not be corrected for in the DAC filters by an anti-ripple of the same shape/structure but 180 degree out of phase?

Link to post
Share on other sites
Nikon Capture NX has also lens profiles and I believe some others too. My Pentax DSLR also has built-in profiles for their own lenses.

Actually, in Lightroom, there are a huge number of profiles to correct for both geometry and vignetting (light dropoff at the edges).

 

Same is true for color profiles for monitors, you can actually make your own with an appropriate sensor and software.

mini+Roon > dCS Rossini DAC + Rossini Master Clock >

Audio Note Kondo Ongaku > Avantgarde Duo Mezzo G2

system pics

Link to post
Share on other sites
Is this correct? "actual filter ringing only occurs in the transition band"?

 

Yes, it is correct, so whenever content exceeds passband you get filter's fingerprint. If the content doesn't have anything that would exceed the passband there's no effect on the output either. (in detail this is quite complicated matter, but when simplified, this is how it works)

 

With steep filters (usually halfband equiripple filters) isn't there ripple in the passband i.e before 20KHz?

 

Passband ripple is different thing and largely depends on filter design methodology and precision of the filter calculations.

 

Isn't this where the concept of impulse response time-smearing come arising from ?

 

Time smearing happens because transient's spectral content becomes limited by the filter, thus exceeding filter's passband. Faster the transient, more high frequency content it has.

 

If the ideal is that the passband should be perfectly flat, can the ADC passband ripple not be corrected for in the DAC filters by an anti-ripple of the same shape/structure but 180 degree out of phase?

 

ADC passband ripple depends on the particular ADC filter and can and should be corrected if needed before anything else is done with the content. If you take a typical studio album that has been recorded at 96/24 in Protools, then mixed and some software synths and effects added, then at mastering stage Eq'ed and converted to 44.1k RedBook for distribution, it is futile to expect to do anything about what happened at ADC stage. At that point you can only deal with ringing introduced by the mastering stage conversion from 96k to 44.1k.

 

Even if distributed at original sampling rate, any level adjustments in digital domain at mixing stage after ADC will remove possibilities of correcting ADC errors like passband ripple, because the scale of the ripple becomes unknown and mixed. So you can only "deblur" using apodizing filters, because this property doesn't really depend on further processing employed on the data between ADC and release.

 

 

(naturally DSD doesn't have these issues... :) )

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to post
Share on other sites
Yes, it is correct, so whenever content exceeds passband you get filter's fingerprint. If the content doesn't have anything that would exceed the passband there's no effect on the output either. (in detail this is quite complicated matter, but when simplified, this is how it works)
Is what you are describing not the Gibb's effect?

 

Passband ripple is different thing and largely depends on filter design methodology and precision of the filter calculations.
Isn't ripple in the passband not a well known issue in signal processing filters & seen in almost all DAC datasheets?

 

Time smearing happens because transient's spectral content becomes limited by the filter, thus exceeding filter's passband. Faster the transient, more high frequency content it has.
Well, yes, any bandlimiting (not just digital filter bandlimiting) will cause a time smear but is there not an additional issue with digital filter ripple also causing more impulse smearing?

 

ADC passband ripple depends on the particular ADC filter and can and should be corrected if needed before anything else is done with the content. If you take a typical studio album that has been recorded at 96/24 in Protools, then mixed and some software synths and effects added, then at mastering stage Eq'ed and converted to 44.1k RedBook for distribution, it is futile to expect to do anything about what happened at ADC stage. At that point you can only deal with ringing introduced by the mastering stage conversion from 96k to 44.1k

 

Even if distributed at original sampling rate, any level adjustments in digital domain at mixing stage after ADC will remove possibilities of correcting ADC errors like passband ripple, because the scale of the ripple becomes unknown and mixed. So you can only "deblur" using apodizing filters, because this property doesn't really depend on further processing employed on the data between ADC and release..

Perhaps, it is futile & it might explain the reports coming in that seem to show MQA processing in the DAC is not as audibly significant as MQA used from ADC to DAC. I guess this would have been a useful question to ask Bob Stuart about MQA?

 

(naturally DSD doesn't have these issues... :) )
Sure
Link to post
Share on other sites
Is what you are describing not the Gibb's effect?

 

Related, characteristics depend on filter's characteristics. So you get sort of fingerprint embedded to the data.

 

Isn't ripple in the passband not a well known issue in signal processing filters & seen in almost all DAC datasheets?

 

Yes, in many DACs, but that is something you can overcome with software implementation... Here's one example which still shows a bit too much due to plotting inaccuracies, Y-axis is in dB:

filter-ripple-28.png

So depends whether you consider that an issue or not (< 0.000000001 dB)...

 

To me, much bigger issue that I prefer to overcome with software filters is inadequate output rate. Most DACs chips use 352.8/384k digital filter output rate, while I prefer to use >10 MHz output rates. So no issues with images and their intermodulation at filter sampling rate multiples...

 

And what is even more issue are not so great sigma-delta modulators inside DAC chips, so better replace those too...

 

Well, yes, any bandlimiting (not just digital filter bandlimiting) will cause a time smear but is there not an additional issue with digital filter ripple also causing more impulse smearing?

 

Passband ripple doesn't itself create any timing smear, it is only frequency response effect, so you get some amount of frequency response alteration as function of frequency. But depending on case it is indicative of rounding/truncation errors in the filter processing which is part of the smearing problem.

 

Perhaps, it is futile & it might explain the reports coming in that seem to show MQA processing in the DAC is not as audibly significant as MQA used from ADC to DAC. I guess this would have been a useful question to ask Bob Stuart about MQA?

 

There is no point in something "from ADC to DAC" unless you mean DRM. You can do correction for ADC and output some standard format and then do correction for DAC. The two are completely separate topics and not connected in any way.

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to post
Share on other sites
Actually, in Lightroom, there are a huge number of profiles to correct for both geometry and vignetting (light dropoff at the edges).

 

Same is true for color profiles for monitors, you can actually make your own with an appropriate sensor and software.

 

That's the point. You would expect these profiles to have the most effect with equipment that's the most out of "alignment" e.g. I see MQA helping the most for less expensive hardware but these principles and techniques are widely known and used for many decades. There's nothing preventing us from developing our own profiles of DACs, amps & speakers using currently available software.

 

Same for printers, you can either painstakingly make your own profiles of printers and papers or use pre canned ones.

 

A real comparison would be between MQA and a properly profiled & corrected system. Not knowing how good the MQA profiles are, hard to say what the results would be. From prior experience in other domains (math is math) I expect that a properly done custom profile would win ;)

 

Not only that but HQPlayer can apply these on the fly in the DSD domain which is extremely cool :):)

Custom room treatments for headphone users.

Link to post
Share on other sites
Related, characteristics depend on filter's characteristics. So you get sort of fingerprint embedded to the data.

 

Yes, in many DACs, but that is something you can overcome with software implementation... Here's one example which still shows a bit too much due to plotting inaccuracies, Y-axis is in dB:

[ATTACH=CONFIG]24077[/ATTACH]

So depends whether you consider that an issue or not (< 0.000000001 dB)...

Sure but the original statement "actual filter ringing only occurs in the transition band" only confuses matters, & is incorrect, IMO

 

To me, much bigger issue that I prefer to overcome with software filters is inadequate output rate. Most DACs chips use 352.8/384k digital filter output rate, while I prefer to use >10 MHz output rates. So no issues with images and their intermodulation at filter sampling rate multiples...
Why do you feel such a high output rate is required?

 

And what is even more issue are not so great sigma-delta modulators inside DAC chips, so better replace those too...
Right

 

Passband ripple doesn't create any timing smear, it is only frequency response effect, so you get some amount of frequency response alteration as function of frequency.
But frequency & time are orthogonally related, no?

 

There is no point in something "from ADC to DAC" unless you mean DRM. You can do correction for ADC and output some standard format and then do correction for DAC. The two are completely separate topics and not connected in any way.
I mean using MQA in ADC recording & using MQA in DAC playback, seems to provide an audibly improved playback, from the small number of reports in so far
Link to post
Share on other sites

I mean using MQA in ADC recording & using MQA in DAC playback, seems to provide an audibly improved playback, from the small number of reports in so far

 

Certainly, at least on those awesome Cookie Marenco recordings ... oh wait ;)

Custom room treatments for headphone users.

Link to post
Share on other sites
Sure but the original statement "actual filter ringing only occurs in the transition band" only confuses matters, & is incorrect, IMO

 

No, we are talking about two entirely different things. Ringing of the step response in time domain vs frequency response ripple in frequency domain!

 

In this example case, the passband ripple is much lower than what 32-bit PCM can represent. So no pass-band ripple in my example case would reach the DAC.

 

Why do you feel such a high output rate is required?

 

To avoid those nasty images around multiples of the digital filter output rate that cause intermodulation products in audible band.

 

For example 0-22 kHz sweep through TI/BB DAC chips used in iFi and Meridian DACs with PCM input:

iDSDmicro-sweep-wide-std.png

 

vs the same but run through proper oversampling filter to 22.2792 MHz and then modulated to DSD:

iDSDmicro-sweep-wide-dsd512.png

 

But frequency & time are orthogonally related, no?

 

Yes, but compare the magnitude of things. For example the ripple is even at worst case below 0.1 dB while the filter attenuation is typically at least 50 dB or so.

 

If you have RedBook content, in normal cases the pass-band ripple is so low that the 16-bit resolution is not able to represent it. But yes, with 24-bit content you could theoretically have it, in case your recording has noise floor lower than 20 bit resolution or so.

 

MQA solves the passband ripple issue without any processing for that purpose because it cuts input to 17-bit resolution... :D

 

Of course it is still better to use filters that don't have passband ripple even within a theoretical 32-bit input resolution of the DAC... ;)

 

 

Given that even best loudspeakers have passband ripple of about +-2.5 dB and even in well acoustically room about +-5 dB, doing digital room correction for those will certainly have huge positive effect, compared to MQA that is designed to prevent use of such...

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to post
Share on other sites

I noticed one thing today. MQA says that .mqa sounds better than Redbook, ok... 2L offers various files for testing that out. All real HiRes files are encoded by Merging, .mqa is encoded with specific MQAencode ver 1.1 and Redbook... is encoded with Lavf55.9.100, known also as FFmpeg, full of weird bugs and famous of low quality.

Exception is track 2L-038SACD, where Redbook is also encoded with MQAencode ver1.1 - don't ask why...

Sorry, english is not my native language.

Fools and fanatics are always certain of themselves, but wiser people are full of doubts.

Link to post
Share on other sites
I noticed one thing today. MQA says that .mqa sounds better than Redbook, ok... 2L offers various files for testing that out. All real HiRes files are encoded by Merging, .mqa is encoded with specific MQAencode ver 1.1 and Redbook... is encoded with Lavf55.9.100, known also as FFmpeg, full of weird bugs and famous of low quality.

Exception is track 2L-038SACD, where Redbook is also encoded with MQAencode ver1.1 - don't ask why...

 

I just take the DXD file and make my own conversions to any other format as necessary... :)

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to post
Share on other sites

Yes, obviously :).

I wanted to show how introducing MQA is not fair for those who can't use DXD and maybe don't know anything about it, but as someone says "MQA is better than anything" they want to listen this out and got somewhat wrong picture...

I don't believe that Merging don't have a option for 16/44 output mode.

Sorry, english is not my native language.

Fools and fanatics are always certain of themselves, but wiser people are full of doubts.

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now


×
×
  • Create New...