Rigor in tracking low level measurements involving PCs

[BE718]

I thought this thread was about low level measurements & don't understand why you want to exclude this topic?

Its not a case of wanting to exclude it.It just seemed a somewhat specific discussion which has a lot more to it than measurement. I thought it should have its own thread, which I was just about to start. Anyhoo, if you feel it should be discussed here I will continue. Mods please move if you feel the following is inappropriate for this thread.


Now, As mentioned I have devised a simple test. We can argue the toss afterwards about its efficacy and relevance, but if you are prepared to humour me I think it will be interesting.

I have compiled 7 test tracks. They are all the same music excerpt. They have been modified in different ways. Obviously I am not going to tell you how All I would like people to do is listen to the files and express what they think the differences are.

Please dont post the opinions here in the thread, I dont want people to influence each other. Please send me a PM with your views.

Next Saturday I will explain all and post the opinions, however I will remove user names so people should feel free to express themselves.

Here is a link to the zip file containing the tracks.

http://gofile.me/2vnEF/XgiWuBHva

Have fun

 

[esldude]

Well don't know how the above files have been made differently. My first thought is to test this distance perception to NFM signals by adding measurable levels of noise floors to see if the level effects depth perception. Frankly it sounds very iffy.

 

[jkeny]

Adding amounts of fixed noise is not the same thing as noise floor modulation - it's the modulation that presumably has an audible effect?

Well don't know how the above files have been made differently. My first thought is to test this distance perception to NFM signals by adding measurable levels of noise floors to see if the level effects depth perception. Frankly it sounds very iffy.

 

[jkeny]

Its not a case of wanting to exclude it.It just seemed a somewhat specific discussion which has a lot more to it than measurement. I thought it should have its own thread, which I was just about to start. Anyhoo, if you feel it should be discussed here I will continue. Mods please move if you feel the following is inappropriate for this thread.


Now, As mentioned I have devised a simple test. We can argue the toss afterwards about its efficacy and relevance, but if you are prepared to humour me I think it will be interesting.

I have compiled 7 test tracks. They are all the same music excerpt. They have been modified in different ways. Obviously I am not going to tell you how All I would like people to do is listen to the files and express what they think the differences are.

Please dont post the opinions here in the thread, I dont want people to influence each other. Please send me a PM with your views.

Next Saturday I will explain all and post the opinions, however I will remove user names so people should feel free to express themselves.

Here is a link to the zip file containing the tracks.

http://gofile.me/2vnEF/XgiWuBHva

Have fun

I think you are right - a listening test should be posted elsewhere. Let's keep the focus of this thread on "tracking low level measurements"
You were mentioning something about some ideas you had for possibly measuring NFM - can you say something about this?

 

[BE718]

Yes, the listening test is a measurement of sorts that may well be very relevant to NFM.

 

[dallasjustice]

This study found that listeners couldn't distinguish various dithers (rectangular, Gaussian, triangular as well as subtractive vs nonsubtractive) at 12 bits. This confirms my skepticism toward the fanciful Chord claims.

Yes, note that it's all about bit rates - the number of bits actually being exercised to encode the signal - which is exactly the same as using a very low level waveform. The remarkable thing is that one can go to quite ridiculously low bit rates, I think I tried just 2 bits at one stage - and my brain could reassemble what the music was about quite adequately, using the right dither - the power of the brain to extract information from noise, if the latter is sufficiently random.

 

[jkeny]

This study found that listeners couldn't distinguish various dithers (rectangular, Gaussian, triangular as well as subtractive vs nonsubtractive) at 12 bits. This confirms my skepticism toward the fanciful Chord claims.

I don't believe we are talking about the same things here - dither randomises the quantisation error at the expense of adding a small level of random noise i.e white noise.

I can do no better than the posts from Opus111 (now Opus112) post before his enforced "holiday from this forum" One of his posts here describes the difference between dither & NFM

"Digital involves quantization. Quantization is a non-linear process that must be dithered to give what appears to be a linear system. Whether its truly linear or only appears to be though is a question I'll not go into right here. We'll stick with the current mainstream view about quantization for now. To get the process to appear linear the dither must be a particular kind - called TPDF for triangular probably density function. The PDF has nothing to do with Adobe, its talking about the signal's statistics, its distribution of levels over time.

Now in a S-D DAC, the quantization isn't at the 16bit level, its much higher. The best (ESS) S-D DACs have only a 6bit DAC. So they involve a quantization process which is throwing away 10bits but then recycling this through a feedback loop so that the noise generated is beyond the audio band. However for this throwing away data to be linear, it must be accompanied by the TPDF dither. I've seen no evidence from the chip manufacturers that this is indeed the case.

What happens when the dither isn't TPDF? Well if you read the theory, the result is noise modulation. RPDF is known to give it for example."

And some posts later he states

In talking about things which get added to the wanted signal, we have two terms - noise, and distortion. Now noise has the characteristic that its something uncorrelated with the signal - it has independence from the music we want to hear. Distortion on the other hand is something which only exists when there's a signal and it has strong correlation to the signal - its at multiples (harmonics) of the signal but only at mathematically well defined frequencies relative to the wanted signal. Its never independent of the signal, always dependent on it. Noise modulation is a third and somewhat neglected category from these other two in that its character is like noise, not harmonics of the signal but it is unlike noise in that its correlated with the signal. No signal, no noise modulation.

As I said, don't get hung up in this -350dB figure - I'm of the opinion that it is just a measure of the performance needed by his noise shaping algorithm to prevent noise floor modulation happening at > -140dB or whatever is the noise floor of Rob's Chord DACs

 

[esldude]

Adding amounts of fixed noise is not the same thing as noise floor modulation - it's the modulation that presumably has an audible effect?

Where did I say anything about fixed noise floors. If we are discussing modulated noise floors and I suggest adding noise at measurable levels I hoped it would be obvious I was referring to modulated noise floors at higher levels. Since it apparently needed spelling out, well now you have it. Add modulated noise floors at measurable levels.

 

[jkeny]

Best to just consider me someone that needs things to be spelled out.

How would you go about adding this modulated noise - noise which modulates with the signal?

Where did I say anything about fixed noise floors. If we are discussing modulated noise floors and I suggest adding noise at measurable levels I hoped it would be obvious I was referring to modulated noise floors at higher levels. Since it apparently needed spelling out, well now you have it. Add modulated noise floors at measurable levels.

 

[esldude]

Best to just consider me someone that needs things to be spelled out.

How would you go about adding this modulated noise - noise which modulates with the signal?

Well, ideally since the fellow at Chord is making the claim I would like to know the type of modulation he has in mind. Knowing nothing one might think the noise floor is modulated by the noise being larger when the signal is larger and noise being lower when the signal is lower. It could also be the spectrum of noise is modulated by the signal. Or something else.

 

[BE718]

JK,

Have you had a chance to download and listen to my test tracks? I know I'm being a bit cryptic but it will help to move this discussion along if people can find an hour to have a listen.

 

[BE718]

I

As I said, don't get hung up in this -350dB figure - I'm of the opinion that it is just a measure of the performance needed by his noise shaping algorithm to prevent noise floor modulation happening at > -140dB or whatever is the noise floor of Rob's Chord DACs

This can be the only sensible conclusion, however we are not interested in the noise shaper per se. if we are talking about effects on the noise floor at, as you say, -140 dB, why does he not talk about those characteristics and effects? The noise shaper may be a cause, but I'm interested in the effect and how it manifests

 

[fas42]

BE718, I've done the download, and have started listening - on a decent laptop, using Media Monkey and Audacity. And there are clearly differences, "better" and "worse" versions. One method I use is to import them all into Audacity, select two, and switch muting between the pair while playing the two - I rank them, and switch their position in the track listing until I get good consistency in rating between adjacent tracks.

Part way through the exercise so far, and still seeing differences between any pair I sample ...

 

[BE718]

BE718, I've done the download, and have started listening - on a decent laptop, using Media Monkey and Audacity. And there are clearly differences, "better" and "worse" versions. One method I use is to import them all into Audacity, select two, and switch muting between the pair while playing the two - I rank them, and switch their position in the track listing until I get good consistency in rating between adjacent tracks.

Part way through the exercise so far, and still seeing differences between any pair I sample ...

Many Thanks, but please dont post findings as I would like not to influence others. Just send me a PM with your conclusions.


Thanks again

 

[fas42]

Yep, will do ...

 

[jkeny]

Downloaded but not listened yet. Can only do headphone listening, at the moment - not great for soundstage differences.

JK,

Have you had a chance to download and listen to my test tracks? I know I'm being a bit cryptic but it will help to move this discussion along if people can find an hour to have a listen.

 

[jkeny]

I guess it might have something to do with marketing -350dB sounds so much more impressive than -140dB

This can be the only sensible conclusion, however we are not interested in the noise shaper per se. if we are talking about effects on the noise floor at, as you say, -140 dB, why does he not talk about those characteristics and effects? The noise shaper may be a cause, but I'm interested in the effect and how it manifests

 

[BE718]

Any responses to the files I posted?

 

[fas42]

Give me a bit more time if possible ... I've had a run of solid days over the last week, and feeling a bit buggered - not quite in the hair splitting mood at the moment for careful listening ...

Cheers,

 

[BE718]

Of course no problem, this is only a fun hobby its not a job