MQN testing/experimentation thread

[jkeny]

Good work John, interesting results.

As you say, the differences are there but are an order of magnitude smaller than one might reasonably need to claim cause and effect.

Indeed what we might be seeing is correlation rather than causation. Still even correlation would be a step in the right direction - giving us some way to measure improvement. I figure that only a body of tests all showing this correlation is the only way to be assured of a correlation.

I have made a first pass at comparative sampling of the dac output using music files and looking for differences. It is proving harder than anticipated to get consistent data from Audacity, so I won't be drawn at this stage. But I have a test setup, and can capture data. A small start, and I need more time to get to grips with the methodology and techniques.

Mark

Yes, I tried Audacity but only briefly as I felt all it could reveal were noise/freq/amplitude differences & this test seemed to me to be more likely to reveal timing differences.

I suspect that both ITD (interaural Timing Differences) & IID(interaural Intensity Difference) cues at play here. The psychoacoustic effect of both together seem to enhance our localisation perception. In nature we use both the timing differences & intensity differences between our ears to locate the source of a sound.

If only one is present we may not analyse it as significant unless above a certain threshold. When both are present in an accurately realistic way we are more sensitive to the effect. I'm sure that by reducing the noise floor we can hear more of the IID cues. And we premise that noise floor reduction is also part of what MQN does. The ideal test would be to have both ITD & IID mapped for the same signal so we could correlate the two but this is currently beyond any test method that I know of. I've always felt that existing audio tests are much too limited as they do not take into account this aspect of our current model of hearing - tending to be limited to one channel tests with simple test signals.

Of course, the kicker in all of this is that a lot of recordings that we listen to are studio generated (do recording engineers go to the bother of panning using psychoacoustically accurate methods when emulating sound stage?) & often not recordings of a live event. Even if they are recordings of a live event we are reliant on the miking techniques used. Anyway, thats a whole udder discussion but worth bearing in mind.

[jkeny]

Another point - even though the individual timing differences between plots are small (maybe only 0.2mS) when we look at any one frequency point, I have found some papers that seem to suggest if more than one frequency is involved, we are more sensitive to ITD. I could be clutching at straws here but if we have a 0.2mS timing difference across the full frequency range does this add up to a more audibly noticeable localisation of the sound?

Jeez, there are so many different possible aspects to this - anybody interested in a Ph.D? :)

See this paper which relates to the barn owl - I'm not sur ethat it states exactly what I said above :) http://www.jneurosci.org/content/6/12/3413.full.pdf

When a sound contains a broad range of frequencies, the owl
localizes the source precisely. When a sound is composed of
only 1 frequency, the owl errs in a manner that suggests that its
auditory system is unable to match unambiguously a given cycle
of the sound in one ear with the corresponding cycle in the other
ear.

[LowOrbit]

Of course, the kicker in all of this is that a lot of recordings that we listen to are studio generated (do recording engineers go to the bother of panning using psychoacoustically accurate methods when emulating sound stage?) & often not recordings of a live event. Even if they are recordings of a live event we are reliant on the miking techniques used. Anyway, thats a whole udder discussion but worth bearing in mind.

I've never paid much attention to "soundstage" or "depth" for precisely these reasons. I spent enough time in recording studios to believe that recording engineers are generally not too concerned (unless they are specifically chasing an ambient aural feel). Studio-recorded albums relay on artificial reverb processes which are not that sophisticated - even the modern convolution processors which impose a "real" room impulse on the reproduced sound are very limited and whilst they sound nice, they have no psychoacoustic dimension to them. And live recordings generally are produced under enormous time and cost pressure and the engineer is happy to get a clean recording and ambience is limited to a few spare mikes dotted about to capture something to drop into the mix to give the feeling of space without any correlation to the direct sound.

[jkeny]

Of course, the kicker in all of this is that a lot of recordings that we listen to are studio generated (do recording engineers go to the bother of panning using psychoacoustically accurate methods when emulating sound stage?) & often not recordings of a live event. Even if they are recordings of a live event we are reliant on the miking techniques used. Anyway, thats a whole udder discussion but worth bearing in mind.

I've never paid much attention to "soundstage" or "depth" for precisely these reasons. I spent enough time in recording studios to believe that recording engineers are generally not too concerned (unless they are specifically chasing an ambient aural feel). Studio-recorded albums relay on artificial reverb processes which are not that sophisticated - even the modern convolution processors which impose a "real" room impulse on the reproduced sound are very limited and whilst they sound nice, they have no psychoacoustic dimension to them. And live recordings generally are produced under enormous time and cost pressure and the engineer is happy to get a clean recording and ambience is limited to a few spare mikes dotted about to capture something to drop into the mix to give the feeling of space without any correlation to the direct sound.

Yep, good point - so I'm not sure where that leaves my theory?

[Diapason]

Yep, good point - so I'm not sure where that leaves my theory?

The classical listeners are still here! ;)

John, I know very little and assume even less, but eyeballing the graphs has me wondering if you can reproduce the results for each individual player. So, if you run the MQn test multiple times are you getting the same graph every time? I'm sure you've thought of this already.

I'm asking because the differences "seem" very small and I'm wondering if they're basically random, or within the realms of experimental error. If you get identical graphs every time you run the test for the same player then obviously I can start to forget about that concern.

[jkeny]

Here are some further plots I did some time ago - pretty sure everything is exactly the same but this time using a 24/96 input file

Foobar:


And this time using Jplay Extreme (older version)


But one thing to note is the generally lower number of timing fluctuations (spikes) in both these plots compared to the 16/44 earlier plots & that Jplay has fewer spikes than Foobar

Again I have focussed on this psychoacoustically sensitive (localisation) area of the frequency spectrum (700Hz - 3KHz) but there may be more important differences at other frequencies? It would be great if some people wanted to contribute to this effort. All that's needed is to take the plot files & use a program which allows you to zoom into various parts of the spectrum to plot the results. In other words to analyse the results for differences. I have already done the recording & generation of these files so no effort required there - just using eyeballs & common sense to spot possible differences between plots

[jkeny]

Yep, good point - so I'm not sure where that leaves my theory?

The classical listeners are still here! ;)

Yea, I was aware of that one & thought I'd leave it for you guys to comment on - classical is dependent on micing technique, I guess & maybe some studio work but I don't know enough about the recording process in general?

John, I know very little and assume even less, but eyeballing the graphs has me wondering if you can reproduce the results for each individual player. So, if you run the MQn test multiple times are you getting the same graph every time? I'm sure you've thought of this already.

I'm asking because the differences "seem" very small and I'm wondering if they're basically random, or within the realms of experimental error. If you get identical graphs every time you run the test for the same player then obviously I can start to forget about that concern.

Yea, I tested repeatability & I will include one of those repeatable plots in my next post to give you a feel for it.

[Diapason]

It would be great if some people wanted to contribute to this effort. All that's needed is to take the plot files & use a program which allows you to zoom into various parts of the spectrum to plot the results. In other words to analyse the results for differences. I have already done the recording & generation of these files so no effort required there - just using eyeballs & common sense to spot possible differences between plots

I may have missed them, but have you got link to the raw data behind the plots? I wouldn't mind having a gander. It'll make a change from looking at currency graphs!

[jkeny]

It would be great if some people wanted to contribute to this effort. All that's needed is to take the plot files & use a program which allows you to zoom into various parts of the spectrum to plot the results. In other words to analyse the results for differences. I have already done the recording & generation of these files so no effort required there - just using eyeballs & common sense to spot possible differences between plots

I may have missed them, but have you got link to the raw data behind the plots? I wouldn't mind having a gander. It'll make a change from looking at currency graphs!

So here are 2 plots of MQN - each one exactly the same conditions & done one after another i.e the same file played back & recorded twice



As you would expect, there are differences & these would be magnified if I zoomed in closer & disappear if I zoomed out more. So what I'm trying to establish is if there is a trend in the plots that correlates with better sound & what is the correct view of the graphs - hence the request for more eyeballs & brains?

As you say & I agree, the spikes in the plots are mostly below 1 microsecond which is insignificant psychoacoustically (it seems). But 2 questions come to mind - the rising slope of timing slippage seemed to be significant - going from 0.2uS at 700Hz to close to 2uS at 3KHz - the ideal would be a flat horizontal line in other words a constant delay across all the frequencies plotted. I'm wondering what audible effect this rising delay has? So if we get a slope closer to a flat line - have we got an audible better sound? The second thing I wonder is, even though these timing fluctuations are small, if the volume of them along the frequencies translate into an audibly fuzzy sound - in other words, taken individually they would not be noticeable but accumulated together along a band of frequencies, would they be more noticeable as a general better focus to the sound? One noticeable difference in these 16/44 plots to the above 24/96 plots is the number of fluctuations being very much lower in the 24/96 plots. Is this an anomaly of the technique or is it a result of the better timing characteristic of 24/96? Is this why we generally find 24/96 sounds better? I would like to try this with DSD but would need to generate my test files in DSD form & play back through the Meitner, for instance. One of the reasons put forth for DSD sounding better is it's better timing characteristics. I haven't listened to DSD Vs PCM on a good system yet. Despite DSD's technical flaws, this timing may trump everything else when it comes to audibility?

As you can see I'm struggling hard to find some correlation between the plots & what we hear :)

I can upload the datafiles but you will also need the scientific plotting program Veusz http://home.gna.org/veusz/
It's a powerful program which I haven't even scratched the surface of, yet. I'm hoping that more can be done with it than I'm currently doing. But for simple plotting it's not daunting - simply open the data files I provide & you can change the axis X (frequency) or Y(Timing) to focus on specific areas of the data.

Loading the folder of results to Google drive & will post the link when it's finished

[sbgk]

Why mqn sd ? I think sq is the best and others report sw as being good, would be interesting to know how these fair in comparison and whether sq does have better timing.