MQA, audio origami and 24/192 music

Audio Gear Talk MQA
1

https://people.xiph.org/~xiphmont/demo/neil-young.html

The music origami video on MQA shows significant signals above 20kHz.
According to the article above these signals should be removed and then the audible spectrum is sampled at 44.1kHz.
If it isn’t removed before sampling it would conflict with and overlap the audible spectrum.

The point of high frequency audio is that the ear is timing sensitive and is affected by signals above 20kHz.
192kHz gives about the same timing resolution as the timing resolution of the ear if I remember correctly.

In one corner sits standard redbook recordings and in the other corner master quality.

It seems like master quality is only better if there is significant sound signals above 20kHz which affects the sound.
The comparison to visible spectra in the article is rather flawed.
If you listen to a live orchestra the audio is not limited to 20kHz audio above 20kHz help you separate different parts of the music.

If you record instruments it would be pretty silly to remove parts of the sound just because you think it is garbage and you only need to sample 44.1kHz. Things would be different if the sound never reached above 20kHz.

I just read the article and found a need to defend 24/192 music.

The goal of this thread is to determine how hi-res music is better not discussing how it is redundant compared to redbook.

2

Haha! Another interesting topic trying to ‘generalise’ that we don’t need Hi-Res recordings. The professionals/recording studios have done Hi-Res recordings long time ago and if we continue to hear the so call ‘down sample’ to 44.1kHz then indeed we are going back to the day when CD was invented!

3

The article above is, in my opinion, simply rubbish. At the very best, it predates the new paradigm ushered in by MQA.

1 Like
5

Sure, wasn’t gunning for you; I’m just disdainful of that article which has surfaced in more than one place recently.

And Bob Stuart says that it can’t; that the ear is far more sensitive to the envelope of the waveform (timing) than what we can hear (frequency). And, in fact, you need >192k to capture this using conventional sampling.

Again, I think that there will be disagreement here. We probably need at least 18-bit.

6

Think he was targeting things regarding PONO player and without the research behind MQA we would probably not understand why we need at least 192kHz rate.

Steve Jobs died 2011 so this is really old news. I brought it up because roon users are passionate about audio.

He’s moving around the 16-bit recording freely making the noise floor equal zero volume or something along those lines.

7

Sorry, but this is wrong on so many levels.
Please read a good paper about how our ears work and what happens in the ear and what happens in your brain. And then come back and tell me again why frequencies beyond 20kHz are so important.
And believe me acoustic is a well explored field of science with no unexplained mystic holes.
Such articles are pseudoscience like astrology or homoeopathy only targeting on getting your money with no real scientific reliability.
If you are not sure ask your otologist.

1 Like
8

I’m placing you in an environment not bandlimited to 20kHz.

There is a massive difference between listening to an recording than listening live.

I honestly don’t know enough to say why frequencies above 20kHz matters but MQA has done research on the topic.

There is clearly a significant part above 20kHz and removing it damages the sound.

9

As the homoeopathists have and you know that it’s only placebo effect

And in this point I must disagree. Please inform yourself about the way the human ear works and you will see what I mean.

10

I think that part of the issue here is that the typical explanation of digital signal processing is vastly over-simplified, and therefore completely misunderstood by a majority of the population. It’s easy to take the sample rate and translate that into some frequency and then relate that back to human hearing and while there’s some validity to that it’s a minor part of the whole story.

In general, bit depth defines dynamic range and sampling rate defines frequency response. These are two fairly simple concepts to understand, but only tell a part of the story.

In reality both sample rate and dynamic range (bit depth) are interrelated. In fact one commonly used method of increasing dynamic range is to increase the sampling frequency before feeding the bitstream to the DAC. In many cases this is combined with a reduction in bit depth so as to overcome some of the issues with being able to resolve 24 bits in the conversion stage.

So, having extended sampling rates can help improve the overall dynamic range of the conversion and this is a good thing.

The larger issue, which is typically ignored, is what happens at the Nyquist frequency (22.05 KHz in Redbook). It’s simple say that everything above that frequency is cut-off by the filter and only the audible goodness remains. If only reality were simple!

There are two ways to apply a filter (analog and digital) and given that we theoretically have 2.05KHz in which to operate analog is pretty much out. This would have to be a very steep filter to avoid dropping into the audible band and that’s nearly impossible to create in the analog domain due to the difficulty in matching components and dealing with thermal drift.

This leaves filtering in the digital domain and while it’s easier to accomplish a steep filter there and massage the hell out of the signal you do this at the expense of phase anomalies and other nasties (like ringing). These errors (noise) are pushed well into the audible band by the filtering process and are one of the reasons that early digital had a reputation for being harsh. Oversampling in the DAC addresses this by moving the ugly digital filter well outside the audio band and allowing for a very gentle analog reconstruction filter at the DAC’s output.

So this is all well and good, but what does it have to do with high resolution audio? Studios (usually / hopefully) process at very high resolutions so as to have the needed headroom in the digital domain in which to mix and EQ. This allows for some information to be destroyed without having an impact on the critical data needed for analog reproduction. When they downsample to 16/44.1 to cut the CD one of the last stages is a digital filter to cut out any information in the digital domain which can fold down into the audible range. This filter is subject to the laws of any other and can have a definite impact on the phase coherency of the information in the audio band.

If the studio instead downsamples to 88.1KHz or 176.4KHz that filter is very far away from the audio band and has much less of an impact on the signal that we’re actually trying to hear.

So, whether or not signals over 20K are interpreted by our brains is largely irrelevant. The bigger issue is that while digital is “just ones and zeros” there’s some pretty involved mathematics at work to make it work. While it’s simple to say that the filter cuts off above X frequency the harsh reality is that the filter is also (and often significantly) mucking with the signal at lower frequencies as well.

But then nobody likes math and it’s so much easier to speak in broad generalizations…

7 Likes
11

After a few years beyond University my memory fades. Especially if you are not doing similar things for a while.

I got the impression you needed a steep digital filter which usually affect phase. If you output analog signals at the sampling rate it is quite obvious you need an analog filter to remove all the high frequencies.

I was also curious if speakers can reach 50kHz which is where MQA places top of section B.
With 192kHz sampling the sound information <20hz is sampled four times as often as 48kHz. Should improve timing and precision.

12

In the DAC side, the bottleneck of 44.1/48k is it needs to go through a more steeper over sampling digital filters; a total of 3 over sampling digital filters each at 2x speed (assuming most modern DACs uses 8x over sampling digital filters) thus this brings the final output stream at 352.8/384k.
If now the input sampling is 352.8/384k, then it simply bypass all the over sampling digital filters and final stage gets converted to analogue. A gentle LPF is used in the analogue output stage. In the case of 44.1/48k will suffer more ‘ringing’ due to the steeper digital filtering. This may explain why hi-res will always sound better due to less aggressive digital filtering at the DAC side. The other aspects in the ADC side, hi-res captures more information above the 20kHz. Therefore, hi-res recording has two main benefits, at the ADC side, it captures more information above the 20k, at the DAC, less aggressive over sampling digital filter that causes the ugly ‘ringing’. In the case of 352.8/384k (DXD) recordings this represents the best PCM playback available today.