advertisement


MQA part the 3rd - t't't'timing...

Jim Audiomisc said:
OK, sorry about the delay. Been a bit distracted by other things and the diagrams took longer to sort out than I'd intended.

The page is now up at
http://www.audiomisc.co.uk/MQA/OnImpulse/RingingInArrears.html

As usual, there will be typos, etc. But it should be readable.
The point wrt 'pre ringing' is that one of MQA's claims seems to be that they 'correct' the ADC with an implict assumption that pre-ringing is a 'bad thing'. That is questionable from basic IT sampling theory and also odd in practice. Hence I wanted to look at the matter with more than one factor involved... which took time. (Certain irony, there. :) )

OK, you can now argue on the basis of what I've actually written. I'm off for a cuppa and some cherry+sultana cake.

BTW I recommend anyone new to the pages I've written about MQA to read the others as well for the sake of context and so I can avoid being asked about things I already covered. Not yet added in all the crosslinks, but the ones present would give a leadin.
Click to expand...
Interesting read, as always.

I do have a basic question,. You start your article with two impulse responses, one is a recognizable "standard" filter type, the other is also recognizable "apodizing" type, now often present as a user choice in modern DACs. The latter is associated with Meridian and MQA.

My question is whether the latter filter response is the result of your previous experiments where you successfully tricked an MQA encoder to decode and render an LPCM impulse as if it was MQA encoded (the aptly named "freewheeling" behavior). Alternatively, is it simply an illustration of a generic filter type, and isn't meant to be a correct representation of actual system behavior?

It's an important question, as your article goes to some lengths to "debunk" potential benefits of the asymmetric filter type, but the fact that the initial figure represents actual MQA end-to-end impulse response remains uncertain. Bob Stuart did state that the encoder, decoder and renderer are meant to work together with a convolved response better than each part - or so I understood.

MQA did recently publish what they claim is their full encode-decode-render impulse response and it's essentially state of the art. One may claim that they are lying, but it's worth noting the substantial lack of alignment with their assertion and your "input" information.

I also hope that you have/will be able to reach out to 2L, so you can remove the "IF" from your ending paragraph:

The main snag for listeners will be, however, that if MQA prevent people from also listening to what went into the MQA process they will be unable to actually tell what – if any! – audible change ‘correcting’ for the ADC may have made.
 
Thanks, Jim. This crystallises a lot of points in various discussions which I have at most half-understood.
I hope to get as far as 75% understanding next time I read.
On a trivial level, the following typo (which I hope was deliberate) made me smile in the context of causality:
“ Indeed, we need one that converts an input impulse into a since pattern. “
 
DimitryZ said:
Interesting read, as always.

I do have a basic question,. You start your article with two impulse responses, one is a recognizable "standard" filter type, the other is also recognizable "apodizing" type, now often present as a user choice in modern DACs. The latter is associated with Meridian and MQA.

My question is whether the latter filter response is the result of your previous experiments where you successfully tricked an MQA encoder to decode and render an LPCM impulse as if it was MQA encoded (the aptly named "freewheeling" behavior).

MQA did recently publish what they claim is their full encode-decode-render impulse response and it's essentially state of the art. One may claim that they are lying, but it's worth noting the substantial lack of alignment with their assertion and your "input" information.

I also hope that you have/will be able to reach out to 2L, so you can remove the "IF" from your ending paragraph:

The main snag for listeners will be, however, that if MQA prevent people from also listening to what went into the MQA process they will be unable to actually tell what – if any! – audible change ‘correcting’ for the ADC may have made.
Click to expand...

The asymmetric response I used as an illustration came from my test inpulses tailed and played via the Meridian Explorer 2 MQA DAC. However I got much the same from GOs files.

'apodising' isn't a good term for this because all decent FIR filters will be 'apodised'.(1) The key distinction in terms of impulse response is the asymmetry which removes 'pre ringing' and presumes it is something we need to avoid.

You'd have to point me at the info you say MQA have put up about their "state of the art" response. It perhaps depends on what you mean by the phrase given that the encode/decode cycle deliberately includes filter details that alias in order to 'encode' and 'decode' the HF.

(1) OK, Robin Watts is trying hard to have a filter as long as the input stream, and thus not apodise at all. 8-] But he is probably defeated in reality because the ADCs will, I suspect, probably be FIR apodised.

I do plan to contact 2L at some point.
 
adamdea said:
Thanks, Jim. This crystallises a lot of points in various discussions which I have at most half-understood.
I hope to get as far as 75% understanding next time I read.
On a trivial level, the following typo (which I hope was deliberate) made me smile in the context of causality:
“ Indeed, we need one that converts an input impulse into a since pattern. “
Click to expand...

If it had been deliberate I might have risked the joke of using 'sincere'. But nope. one of typooos. Hope to sqush some later today.
 
Looking with a fuller browser I'm not clear on some points. Most obvious one is that the signal plots give the times in a format I'm not sure I recognise. Give values like "00:00:04:05 and says "75 fps". How does that translate into second or millseconds, etc? Not knowing, it just looks like the timescale is too compressed to show the details of edges or impulses.

They look 'perfect' which of course you can't have from a finite sample rate unless your 'filter' is just one sample long. But I have no idea what instrument/software he used to plot the pretty results.
 
I've now fixed the typos and tweaked the engerlush, and added the links to other pages, etc. Hopefully haven't missed anything too confusing. You may need to refresh your browser to get the updated version.
 
'apodising' isn't a good term for this because all decent FIR filters will be 'apodised'.
Click to expand...

Have to disagree here.

Apodising in the context of reconstruction filtering has been coined by Meridian with the strict meaning that the filter reaches its stop-band well before Fs/2, thereby supplanting whatever the dominant ADC or mastering anti-alias filter did.

Most FIR filters used in DACs (at least until recently) are half-band and reach only -3 to -10 dB or so at Fs/2. And then the mandatory MQA filter for non-MQA source material does not even try to reach anything at Fs/2 (*).


(* Which puts it in ideological contradiction with Meridian's earlier intentions. Please contrast Explorer with Explorer 2.)
 
Werner said:
Have to disagree here.

Apodising in the context of reconstruction filtering has been coined by Meridian with the strict meaning that the filter reaches its stop-band well before Fs/2, thereby supplanting whatever the dominant ADC or mastering anti-alias filter did.

Most FIR filters used in DACs (at least until recently) are half-band and reach only -3 to -10 dB or so at Fs/2. And then the mandatory MQA filter for non-MQA source material does not even try to reach anything at Fs/2 (*).

(* Which puts it in ideological contradiction with Meridian's earlier intentions. Please contrast Explorer with Explorer 2.)
Click to expand...

The terms 'apodised' and 'apodising' were used *long* before Meridian tried to plant a flag on their alternative use. Should know as I was building and using Polarising Michelson Interferometers and FFTs for years before digital audio or Meridian appeared!

In the context of FIR ADC/DACs for audio, the aim is to tailor the imperfections caused by the chosen finite range, etc. But follows the same maths.

So their attempts just muddy the waters for the sake of being able to impress people with their 'technical' language. Classical language scholars might also have a view... :)

Hmmm... maybe I should do some distortion measurements to compare my ancient Meridian DACs with the Exploder 2.

Is this why the plots on their webpage (as refd earlier) have unfeasably clean squarewaves? If so, beware test squarewaves whose period isn't a neat integer number of samples per half-cycle.
 
Hi Jim, does apodising, in the sense you use it mean the same as windowing?

On a completely separate note, has anyone ever considered, in the context of pre-ringing and its alleged perniciousness, the phenomenon of auditory backward masking. I believe that it is what my kids would call "an actual thing".
 
Yes, apodising is old. But for consumer audio Meridian used it first, in a very specific meaning.

in the context of pre-ringing and its alleged perniciousness, the phenomenon of auditory backward masking.
Click to expand...

Pre-ringing can be perfectly audible: when it happens at audible frequencies, i.e. in the mid-range of the human ear.

At inaudible frequencies ... not.
 
adamdea said:
Hi Jim, does apodising, in the sense you use it mean the same as windowing?

On a completely separate note, has anyone ever considered, in the context of pre-ringing and its alleged perniciousness, the phenomenon of auditory backward masking. I believe that it is what my kids would call "an actual thing".
Click to expand...

It's more a case of 'windowing' being what you *do* to get 'apodisation'. But you can window a data set with some other aim in mind. cf my next comment below
 
Werner said:
Yes, apodising is old. But for consumer audio Meridian used it first, in a very specific meaning.

Pre-ringing can be perfectly audible: when it happens at audible frequencies, i.e. in the mid-range of the human ear.

At inaudible frequencies ... not.
Click to expand...

Yes. they may have done. But it just leads to confusion. Example of using a term to 'bafflegab' the punters. More generally, the term gets applied in various topics. e.g. look in Born and Wolf (Optics) for it applied to modifying the behaviour of lenses of finite size, or in books on antennas where it may be used when discussing the illumination pattern of a dish or the antenna pattern, etc.

However in the context here I think, actually, that it makes more sense to describe the difference between the two impulse responses as being 'symmetric' and 'asymmetric' because more people will know what symmetric means and the difference which is relevant here *is* the choice between time-symmtric and asymmetric.

i..e. in this context 'apodised' simply acts as bafflegab for most readers. Whereas they can *see* the symmetry in the plots! Seems better to me as an aid to understanding.

Yes, I agree about 'pre ringing' being audible if you drop the filter down to lower frequencies. In part because you then get to a long enough, loud enough, 'pre ring'.

But up above 20kHz it seems unlikely that any real audio recordings will cause a problem. Particularly with real audio that matches MQA's '1/f spectrum' demand. :)

And again, I dislike the 'acausal' labels as they ignore the reality that filters add in a delay by their nature.
 
Jim Audiomisc said:
Yes, I agree about 'pre ringing' being audible if you drop the filter down to lower frequencies. In part because you then get to a long enough, loud enough, 'pre ring'.
Click to expand...
In this context it struck me as odd that the question of how loud, how long before etc never seems to come into the disucssion from those who hyothesised that pre-ringing might be "the problem" with digital audio. After all we are talking about impulses here and surely if anything was a candidate for masking this would be it, although obviously it's applicablity to inaudible sounds may be questionable. Perhaps pre-masking is inapplicable to a signkle smeared shape as opposed to two distict sounds. Which seems to me to take us to other topics in the time resoltion of hearing eg how far apart two sounds have to be to be perceived as separate and in turn to...
And again, I dislike the 'acausal' labels as they ignore the reality that filters add in a delay by their nature.
Click to expand...
Also where you input an impulse to an ordinary filter and get an output with a non-zero rise time, then we don't habitually say that the rising edge is acausal just becasue it precedes the moment of peak amplitude do we? So why should we regard the pre-ringing differently. The implicit argument is that you somehow know that the loud bit happened at exactly t=0 and that any signal before t=0 must be the result of some impossible backwards causation. Sed quaere on so many different levels.
 
No-one seems to have responded to my comments/questions in #30 (which refers to the page URL'd in #29, etc).

Does anyone know what the time-scale shown is on the graph in question? I still don't know what the values with multiple ':' dividers mean. I could have understood HH:MM:SS.s but it has too many dividers. All I can deduce is that the scale is too gross to show details of the squarewave or impulses. The plots seem ''impossibly perfect".
 
Jim Audiomisc said:
No-one seems to have responded to my comments/questions in #30 (which refers to the page URL'd in #29, etc).

Does anyone know what the time-scale shown is on the graph in question? I still don't know what the values with multiple ':' dividers mean. I could have understood HH:MM:SS.s but it has too many dividers. All I can deduce is that the scale is too gross to show details of the squarewave or impulses. The plots seem ''impossibly perfect".
Click to expand...
do you mean
https://bobtalks.co.uk/wp-content/uploads/2021/05/882-Square.png
if so the dividers look to be tenths of seconds (marked with the time 00:00:01:07.3 etc) and an unlabelled divider between, presumably equalling 20ths of seconds.
 
You must log in or register to reply here.


advertisement


Back
Top