Temporal Confusion

[Miska]

Maybe my question is whether the slope of a signal has to be steeper than what the audible frequencies can represent to reproduce all the transients?

 

We've discussed the topic several times here on this forum. My take is that yes, that is the case. Especially because brickwall filters used for CD sampling rates have much wider time domain impact area than just the transient steepness change. So essentially the leading slope typically becomes at least as slow as 500 µs. And another 500 µs worth of the trailing slope, but that is usually masked by the natural transient decay of the instruments. But note that this affects only the cases where the filter cuts in, not the cases where you have for example constant 20 kHz tone and nothing for the filter to remove.

[YashN]

MQA first splits the band into below and above 20 khz. It would use 48 khz for each half. One sometimes forgotten idea about Shannon-Nyquist is you can use your bandwith where you want it. For instance you could use 48 khz sample rates to record a 20 khz wide band between 140 khz and 160 khz. So MQA will use 48 khz below 20khz and another 48khz from 20khz to 40 khz.

 

The 20-40 khz band MQA will use lossless compression to cover up to about 30 khz. It will use lossy compression to cover above 30 khz. Unlike additive dither we normally see in audio it will use subtractive dither and noise shaping. It will keep only the difference signal between the original 20-40 khz band and what is in the 20-30khz band. It will encode the info to perform subtractive dither in that lossy code in a way upon decoding it can be retrieved. So MQA will only be lossless to 30 khz and will be heavily compressed though of good resolution due to noise shaping and subtractive dither. Lets it encode potentially a wide bandwidth in only 3 bits. The dither allows it to have enough resolution to reconstitute a good fascimile of the original whole wide bandwidth signal.

 

Very interesting, thanks for that.

 

Stuart has been clear that we don't actually hear the ultrasonics in his opinion. He thinks we do hear filtering. 30 khz and lossy above that up to much higher frequencies will allow gentle presumably non-audible filtering with better time domain performance.

 

The importance of time domain has been my understanding for a good while now. For me, more precisely, attack transients. There is a staggering amount of information in the early milliseconds at the onset of sound. Crucial for location, emotion/feelings/flight response and timbre-recognition.

 

They have said MQA could be used with any original file up to 384 khz. So it sounds like they would do their MQA processing and let you play back the 384 file at 192 I think.

 

That's good compression.

[trithio]

I was reading about Meridian’s MQA and one statement was that it was based on the newest psychoacoustical research but I have not seen any references on what that research is. I anyhow hate claims that something is based on research but without any proof of that. As a scientist I believe that any statement should be verifiable.

Not knowing what Meridian refers to I made some web searches. Almost all of which is not very new research!

1. Frequencies: Humans can hear between 16-24Hz lower limit to 12-24kHz higher limit depending on age, gender, hearing damage. Also dependent on levels see Fletcher-Munson. Bone conduction is often quoted for extended frequencies but I think it is irrelevant for music as it is pitch and pace insensitive at higher frequencies.

2. Dynamic range: there is a lower limit which is frequency dependent again see Fletcher-Munson. The upper limit is time dependent. Long term sound can cause hearing damage already 90dB but short term upper maximum comfort level is at 120dB. Not clear what short term means msec or sec?

3. Time resolution about 10µsec (which would equate 3.4mm sound travel in air). This is mostly import for localization of sound.

4. Detectability of differences (like distortion) is frequency and level dependent. I did not find much about that except https://www.meridian-audio.com/meridian-uploads/ara/coding2.pdf on page 25.

There are a lot of other interesting effects like masking etc.

 

So what would that mean for the digital reproduction of audio?

From 1 and 2: 48 kHz sampling rate and 20 bit should be sufficient. Imagine a single wave form representing that frequency and dynamic range from a single source (like a microphone). A time change of that wave form by 10 µsec should not be audible by itself but when referenced to another wave form like left/right audio channel or other microphone that time shift will become audible according to 3.

So my confusion: Do you need a higher sampling rate than 48/20 to get that time resolution or or as long as encoding and decoding preserves that time resolution 48/20 is sufficient? What about digital filters and pre-ringing?

More of a problem preserving the 10µsec time resolution I see in recording and loudspeakers. Already different microphones and cables will have different LCR values which will cause phase shifts and therefore time shifts >10µsec. Maybe that is the reason I like many simple stereo recordings with only 2 identical microphones and cables. I wonder what can be achieved in terms of time resolution with modern multi-micing, digital mixing and corrections. One could digitally correct for time errors but with what precision is that done? Loudspeakers with multiple drivers and a cross-over have significant time problems. Alignment within less than a msec is difficult if they are at all time aligned.

 

Here are some more or less official db + duration standards for dangerous loudness

http://www.lowertheboom.org/trice/safedblevels.htm

http://www.dangerousdecibels.org/education/information-center/decibel-exposure-time-guidelines/

 

Congrats for the thread, yours was the first CA post that I did not fully understand after a summary read. Which is highly welcome as it's the only time when you learn something.

As about meridian's new format, even if I dont get all the technicalities that thing has all the signs of a marketing driven scheme. Including the good old "based on science" trick presented without any other info.