MQN testing/experimentation thread

[tony]

That looks interesting John. Is it moving towards providing fact based evidence for those that require proof. My basic technical understanding is that it is showing a difference between software players.

[Diapason]

John that sounds like some solid work! Will have a look at that link when I'm back in front of a real computer, but very interesting reading.

[wushuliu]

So I got a new mobo and cpu yet again, this time Haswell G3220 and 1150 MSI board. Partly because it has more overclocking features, partly because the MSI boards claim to be more EMI resistant and higher quality than previous generation. Anyway, I am now able to reduce Ram voltage. I just lowered from 1.5v to 1.425 and there is an audible change in presentation. Anyone else lowering ram voltage and if so how much?

[nige2000]

So I got a new mobo and cpu yet again, this time Haswell G3220 and 1150 MSI board. Partly because it has more overclocking features, partly because the MSI boards claim to be more EMI resistant and higher quality than previous generation. Anyway, I am now able to reduce Ram voltage. I just lowered from 1.5v to 1.425 and there is an audible change in presentation. Anyone else lowering ram voltage and if so how much?

this stuff is probably best in build a audio pc thread,
try ram at 800 mhz and 1.25-1.4v
but really should start with supplying clean power, rather than trying to sort out a dirty supply
linear supply and pico psu are good and are reusable if you want to go further

[jkeny]

OK, first I fixed the image display in my post so images now appearing in-line which makes it all much more understandable, I hope.

Before doing anymore tests (yes testing different versions will be interesting) we need to tease out what is being looked at, what the dependencies are, what the limitations are, etc. Hopefully this can focus our further tests?

(The MQN version used was 1024 SD for reference)

This is a new test so we have to be careful with analysing the results & I'm hoping you guys can help in this?

Let me explain what the test does - it generates a file with two channels. One channel is a Sine wave & the other channel a Cosine wave. What this means is that for any digital sample point in the Sine wave channel there is a known mathematical relationship with it's corresponding Cosine wave digital sample point - call these two points a sample pair across the channels.

When the file is processed through playback software & hardware, if there is any timing slippage, it will be picked up by analysing the output. There is a software program which does this analysis & can show the timing slippage at all the frequencies 0 -22KHz.

So a perfect playback gives a flat horizontal line at zero which means n change in timing at any frequency.

What we see in the real measurements is:
- not a fine line but a fuzzy line which when you zoom is is seen to be individual fluctuations of timing at various frequencies
- this fuzzy line rises from low frequency to high frequency meaning that the timing is getting worse as we go to higher frequencies

Now the questions:
- what causes the individual frequency timing fluctuations? I suspect this is a result of a number of factors - the variations in the computer; variations in the recording hardware; variations caused by the software's affect on the power in the computer. Others I haven't considered?
- what causes the rising timing differences as frequency increases. I suspect this is the hardware i.e the DAC & recorder

So what we are looking for is the timing effects of different playback software overlaid on top of these other factors affecting timing. Tricky, eh? And that's why we have to be careful in our analysis.

Just as an aside - this will never be proof for the objectors as it requires interpretation/analysis & we all know what that means - fight, fight, fight. The best that can be achieved is to build up a body of tests which all show the effects we are talking about.

Looking at the zoomed in 300Hz to 1KHz plots two things (I think) are noticeable:
- the MQN slope of the line is lower than Foobar's line - meaning it has lower timing slippage across these frequencies. I think this is evident across all frequencies so we have a lower latency in MQN (I haven't done any registry changes, btw & all these measurements were done one after another without change to try to ensure as much consistency as possible)
- I could be wrong but it appears as if the amount of individual fluctuations in timing at each frequency is also lower in MQN? This is a harder one to call.

As I said the psychoacoustically sensitive area for sound localisation i.e sound stage is timing in the approx freq range 300Hz to 3 or 5KHz (I'd have to look it up again). Both the differences in timing & the difference in volume between each ear are the two main factors used to generate sound stage (other things such as primary sound Vs reflections also come into play in rooms as opposed to headphones). So there may well be another, completely separate factor operating here & that is that MQN, by lowering the noise floor allows us to better hear the loudness difference cues at each ear - hence a better sound stage illusion is created.

Anyway, I might be out for most of the day (match, etc.) but hope to be able to pick this up later or tomorrow.

[jkeny]

That looks interesting John. Is it moving towards providing fact based evidence for those that require proof. My basic technical understanding is that it is showing a difference between software players.

Doubt it Tony - it's too open to interpretation to be considered as evidence by the doubters.

I'll give you an example of how much effort is needed:
I raised the two's complement padding error on another forum & had two basic types of response:
- one, I was a retard & just a solder jockey - even when I went through the maths this guy tried to use maths to show how wrong the concept was. And this guy is not a fool (in the common understanding of the term)
- another guy, a luminary in psychoacoustic testing, JJ Johnston, started by poo-pooing the idea but then fell back on the defence that audio manufacturers get things wrong all the time - when, in fact, it is the I2S standard that is wrong.

So don't expect this set of measurements to change anything.
What I hope it might help is our testing of MQN with some handle on measurable differences but as you can see the measured differences between Foobar & MQN (1024) are not obvious so little hope of seeing differences between SD & SB of 1024 MQN

[jkeny]

Fran, sorry to bother again but could you hive off the last number of measurements posts & replies to another thread just to keep it separate & to allow the other stream of the thread to continue - experimenting?

[jkeny]

One final comment - we have to be careful not to confuse correlation with causation (before this is pointed out). In other words just because it might show a difference in measurement doesn't mean that this is necessarily the cause of the audible differences we hear - it may be but other tests are needed to verify this.

I still think that different software players probably have a multi-faceted effect - both timing & noise being the two most obvious areas in which they improve matters. Taken separately neither of these may show significant differences but together they both fit our psychoacoustic model of hearing - when there are subtle timing differences in the signal that reaches each ear which are also allied to subtle loudness differences (which are revealed by the lower noise) at each ear, we get a synergy which matches the way we perceive location in the real world - so the sound stage form our playback system appears more realistic.

BTW, this test can be used to evaluate hardware in this area of timing. So a preamplifier or amplifier could also be measured to evaluate & correlate any measured difference to audible difference. In fact this may be a worthwhile area to investigate in order to prove/disprove our premise that timing differences between signals on each channel can have an audible effect. If we could identify a system where the sound stage collapses because of one change & measure the timing on it's output before & after collapse, it might help us get a handle on these measurements? I have ideas about this.

BTW, it might be interesting to test cables with this method - ones that provide better sound stage Vs ones that don't. Does anybody have such examples or are there such cables?
Another thing that could be tested (but this is more difficult) would be speakers. Can they resolve timing down to the 1-5 microseconds needed for good sound stage reproduction?

[jkeny]

The other thing that I forgot to mention is what the frequency means on these plots.
It's the frequency of the timing errors - in other words low frequency means a timing error that has been analysed across many samples - a high frequency error is across fewer samples.

These plots give this view where the x-axis is time & the Y-axis is PPM (parts per million) errors

First Foobar


And now MQN- 1024-sd


Don't really know what to say about these except that MQN's timings look worse than Foobar

[jesuscheung]

try using 2.71 v1 and v2 for test.

those new versions of MQn e.g. sd has 3D effects. very difficult to sound correct.