Noise from computer, how ?

[bdohmeyer]

Hello,

I'm strongly leaning toward going with the XX and Phasure NOS1 route (and build a new computer with recommended components).  Although I am concerned that my complete ignorance of computers will be a significant hindrance (to me and everyone else), Juan has assured me that I should post my queries without fear. 

I have read hundreds of posts to attempt to learn as much as I can.  However, one very important thing is not clear to me.  There is a great deal of discussion regarding computer tweaks generally and noise related to the computer - usb3.0 vs 2.0, fans, power supply etc.

So, since the computer is essentially a transport (after the very cool processing) how can anything that has no effect on zeros and ones effect SQ?

If noise is being transported from the computer to the dac via a usb connection then could we theoretically use an optical filter? (I have no idea what I'm talking about - however, it seems impossible for noise to move down a light driven connection)

Thank you,

Bob Dohmeyer

[PeterSt]

Hi there Bob - warm welcome here.

(please notice that your post plus response(s) to it are candidate to move to another board later - probably Support or maybe Chatter)

First off, you don't want to know how many people with a fair amount of more hindrance than you got all going - but it may require some help which is available here (like you apparently contacted Juan in private). Otherwise we at Phasure have a PC ready for you - get it out of the box, hook up your NOS1, start XXHighEnd and press Play. But this NOT assumed :

Many people go crazily far with the "noise" and such, and you know what ? I myself am even not amongst them. That is, not this PC noise. Software impeded yes, but that's what XXHighEnd is for from top to bottom.
... which does not tell that getting those last pieces of noise out of the way will not improve ...

I will try to give you one example of how noise even "goes through light" - which hopefully explains :

Your zeros and ones - transported from PC to DAC - is essentially an analog signal, with a fixed amount of plus voltage for the ones and (close to) zero for the zeroes. The sequence of the zeros and ones create "data" as such, which is binary code but further not important for our story. The fact that this is just an analog signal is though;

When the signal goes to plus, something has to happen (like interpreting it *is* a plus or better : a one). This going to plus goes with a certain "ramp up" hence steepness. So, not veryically straight up, but with a slope (which is as steep as possible, but inifinitely steep is not possible in electornics). Now :

The devices which interpret this 1-0-1-0 sequence, have a threshold for when a 1 becomes a 1. So, say that the peak value in voltage of this signal is 1.2, then at 0.8 and more it will be interpreted as a one. Under that it should be zero. But now comes the trickery part :

If this would be a completely noiseless signal, the moment a zero becomes a one (or the other way around) would be completely fixed in time. With this I mean : from the one part of the signal to the other, with a virtual evenly space between the "data", from one part to the other goes in a fixed time (think 0-0-0-0 or 1-1-1-1 or 1-0-1-1 etc. and the "-'" parts being this fixed time). However :
When there is noise on the signal, and notice that noise is a "wave" within itself, it disturbes this fixed time in between these parts, and this "jittery" property of the data is, well, jitter !
So, this noise-wave on the signal makes the signal wobble, and because of that the detection of a one can come early or late (detection of a zero just the same).

The means our audio samples are put out from a DAC works via above described means (a voltage goes up -> put out the sample), and is thus subject to jitter once noise is in the signal (there are more sources for jitter, but since you wondered about noise, this is our subject today).

Once the samples are not put out evenly spaced in time, this is audible. Think about the wow and/or flutter of a turn table, and the effect is exactly the same, only it goes 10s of thousands of times faster. Now you can imagine how digital can sound nasty - just envision the "judder" of a turn table to be there infinitely more fast.

So, I started out with your "noise through light", and here is your answer :
The means at the DAC end to "convert" light to voltage again, is a noise source by itself.
The remainder was explained above ...

And for the non-light connections : Any noise which comes from the computer travelling along with the data, implies the same noise at the DAC end.

Easy ? I think so. Once someone explains it, on which I hopefully succeeded somewhat !

Kind regards,
Peter

[AlainGr]

Hi there Bob - warm welcome here.

(please notice that your post plus response(s) to it are candidate to move to another board later - probably Support or maybe Chatter)

First off, you don't want to know how many people with a fair amount of more hindrance than you got all going - but it may require some help which is available here (like you apparently contacted Juan in private). Otherwise we at Phasure have a PC ready for you - get it out of the box, hook up your NOS1, start XXHighEnd and press Play. But this NOT assumed :

Many people go crazily far with the "noise" and such, and you know what ? I myself am even not amongst them. That is, not this PC noise. Software impeded yes, but that's what XXHighEnd is for from top to bottom.
... which does not tell that getting those last pieces of noise out of the way will not improve ...

I will try to give you one example of how noise even "goes through light" - which hopefully explains :

Your zeros and ones - transported from PC to DAC - is essentially an analog signal, with a fixed amount of plus voltage for the ones and (close to) zero for the zeroes. The sequence of the zeros and ones create "data" as such, which is binary code but further not important for our story. The fact that this is just an analog signal is though;

When the signal goes to plus, something has to happen (like interpreting it *is* a plus or better : a one). This going to plus goes with a certain "ramp up" hence steepness. So, not veryically straight up, but with a slope (which is as steep as possible, but inifinitely steep is not possible in electornics). Now :

The devices which interpret this 1-0-1-0 sequence, have a threshold for when a 1 becomes a 1. So, say that the peak value in voltage of this signal is 1.2, then at 0.8 and more it will be interpreted as a one. Under that it should be zero. But now comes the trickery part :

If this would be a completely noiseless signal, the moment a zero becomes a one (or the other way around) would be completely fixed in time. With this I mean : from the one part of the signal to the other, with a virtual evenly space between the "data", from one part to the other goes in a fixed time (think 0-0-0-0 or 1-1-1-1 or 1-0-1-1 etc. and the "-'" parts being this fixed time). However :
When there is noise on the signal, and notice that noise is a "wave" within itself, it disturbes this fixed time in between these parts, and this "jittery" property of the data is, well, jitter !
So, this noise-wave on the signal makes the signal wobble, and because of that the detection of a one can come early or late (detection of a zero just the same).

The means our audio samples are put out from a DAC works via above described means (a voltage goes up -> put out the sample), and is thus subject to jitter once noise is in the signal (there are more sources for jitter, but since you wondered about noise, this is our subject today).

Once the samples are not put out evenly spaced in time, this is audible. Think about the wow and/or flutter of a turn table, and the effect is exactly the same, only it goes 10s of thousands of times faster. Now you can imagine how digital can sound nasty - just envision the "judder" of a turn table to be there infinitely more fast.

So, I started out with your "noise through light", and here is your answer :
The means at the DAC end to "convert" light to voltage again, is a noise source by itself.
The remainder was explained above ...

And for the non-light connections : Any noise which comes from the computer travelling along with the data, implies the same noise at the DAC end.

Easy ? I think so. Once someone explains it, on which I hopefully succeeded somewhat !

Kind regards,
Peter

Since I read all you write, I can say that this post should be put in a general document or in a FAQ thread

It is not the only one, but it is very important and useful

Alain

[bdohmeyer]

Hi Peter,

Thank you so much - that really helped.  So even though the noise is bit perfect, it's off in time and flow. 

Where can I go to better understand this jittter and "re-clocking" issue better?

My rig is vinyl only right now but I'm very excited about what you have here and can't wait to get it gong.

Sorry about the topic change (I tried to tie it to the 2.0 vs 3.0 USB tweak but everyone immediately knew it was out of place)

Bob

[PeterSt]

Bob,

Where can I go to better understand this jittter and "re-clocking" issue better?

I tried to find you something which didn't dive into matters deeply, but I failed. But in case you derived "re-clocking" from my post as a solution to the (noise) issues metioned, I don't think this is much related.
Reclocking merely will be about improving a jittery signal with a better (more precise) clock, where the signal has an inherent bad "jitter figure", for example SPDIF. But it is all too complicated to explain in layman's terms, already because quite some causes of jitter exist, more types of jitter exist, and so many solutions to already one cause/type exist. Besides that, nobody is right because there's actually nothing available to test one solution next to another.
Lastly, all what can be found is quite old by now, since tecniques progress vastly (and fast).

To give an example of how techniques *have* to progress :
You will read everywhere that asynchronous USB can carry no jitter (skip the reason why). But I say that it will induce the most jitter of all kind of connections. And if I am not alone "anymore" on this it is probably because people copied my ideas about it.
It really is a difficult subject.

The NOS1 you may obtain does not perform any reclocking. This is not necessary because all the means to do that will make it worse. Ah, that sounds strange eh ? But in brief it comes down to taking out the noise before it can harm and next have the clocks (oscillators) positioned so that no other means of jitter can be introduced. Of course this sounds like a blah blah story, but it should be proven by the sort of fact that I not only claim that better USB cables won't help, but that also nobody really succeeded in that. So, read around the internet about all sorts of USB cable comparisons and how the one is even better than the other. Not with the NOS1.
Same story with mains cords btw.

I'm not sure where I wanted to go to with the above because the story really *is* difficult. So it is a bit annecdotical although things like "use cheap cords" is promised/claimed in advance by me.
Actually the best proof would be understanding how XXHighEnd operates and how it influences the DAC in - up to a certain extend - controlled fahsion with its dials and buttons. Well, the "how" are my secrets, but the "what" should be a clear thing : it can only be jitter. That is, as long as the zeros and ones don't change (which they don't) there is one single means left - jitter.
And you know how jitter is infuenced ... by noise.

So now the circle is complete, since it was you who wondered how noise from the PC could influence sound quality.
All I did was creating some software influencing the noise for the better, and with that knowlegde created a DAC to emphasize improvements by software, which latter are actually quite marginal in relative sense.

And being in this stage of the story, we still see that hardware improvements in the PC can (still) also matter.
In the far end, when the software "knows" about what all influences and how it can be eliminated, the software can be used to leave the PC alone. But this too is a quite complex story because XXHighEnd is slowly eliminating herself. Confusing ? yes, I can imagine. But accidentally in the latest XXHighEnd versions it has become clear that dials which previously had quite an impact now don't have that so much anymore (this is a positive). And all what happened was the introduction of that new "dial" - a stronger one.

If you read back (which I just did) you see a contradiction in how noise from the USB connection is no issue, while at the same time we can "steer" noise from the PC which obviously has to travel through that same USB connection. So you see ? it is not so easy and would take miles of pages to explain.
And then to think that only a couple of years back nobody believed PC's or software could make a difference to the sound to begin with (outside equalizing / DSP of course).

Hope this helped somewhat, but I think that is doubtful.
Peter

[Jud]

But in brief it comes down to taking out the noise before it can harm and next have the clocks (oscillators) positioned so that no other means of jitter can be introduced.

Does this also keep noise off of ground (at least from being transmitted through or getting into the DAC via ground)?  But then one would have to provide separately for noise coming from the mains into other parts of the system, I assume.

[bdohmeyer]

Peter,

Again, thank you so much for your time and expertise. I was recently researching the details of how a hi rez file's SQ was so much improved and came across a MSB "How DACS work" The explanation of the ladder DAC was a big "ah-ha" moment for me - I was going crazy wondering if it was just me or was CD sounding worse over the years with all the processing.  So I researched ladder DACS and nearly settled on the Metrum HEX until I read the NOS1 6moons review and some comments you made on a CA thread on the need to up-sample for the ladder DAC's (without filter related ringing).

Your comments are really helping - before I believed that your controls changed the sequence of 01's.  

So, hopefully, so I can understand the computer noise / jitter issue without being a big pain in the butt, would we theoretically eliminate the issue if we had a separate device / hard drive that recorded the bit-perfect but time flawed data
from the computer and that theoretical separate device could theoretically replay the data without computer noise / jitter ?.  I understand how impractical this is (without ability to change volume etc) - just asking to understand the issue better.

Thanks again,

Bob

PS - During my research I came across another "ah-ha" moment of another reason why CD seemed to sound worse over the years with processing (especially during loud passages) when it was explained that the processing interpolations could at times extend past the dynamic range of the system and cause "digital clipping".  Is that what peak extension is for?

[PeterSt]

Hi again Bob.

Since I was measuring anyway, here's a small bonus (also for others - see more below) for hopefully better insight :

Although this is a measurement on the outputs of the NOS1, it nicely can show what I meant with jitter because of the 0-1's hammered upon by noise.
You see a slight wave, and just suppose the 11.00ms and 11.50ms marks are to be the "trigger points" for my before-mentioned voltage (detected) change. Try to envision that these both marks are the official time points where the 1's and 0' need to flip. And, with some imagination you can see that the wave has its peaks at 10.75ms and 11.75ms and its dips at 11.23 and 12.25ms (the peaks work out better than the dips here).

Now, because of the noise this will never work out. But now compare with this picture (but do NOT take into account the high peaks you see - they are unrelated to the story) :

So, envisioning those peaks not being there, it is clear that at 0.45ms or so the voltage goes through zero and at ~ 0.95ms it goes up. Same at 1.45ms and 1.95ms.

If you compare the both scales you can see that peak to peak noise in the first picture is around 55uV and in the second picture it is around 40uV.

Because of the fewer noise in the second picture, that will imply less jitter (well, the first picture actually has no chance).


About the bonus for others :

What I did here was abusing the normal output of the DAC to sort of show how noise can let fail all. But what really was shown is the resolution of the 24 bit NOS1 which really won't resolve more than 20 bits because of noise in the output.
To keep in mind : this is perfectly normal.
So, this is 120dBFS of attenuation of a 1000Hz test signal, and attenuate 6dBFS more and the "wave" will disappear in the height of the noise. Notice that attenuating 6dBFS means the shown peak-peak voltage will be divided by two. So, the 55uV you see will become 22.5uV and it is not so difficult to see how that will be as high (or less high) than the high frequency noise. Gone is our signal.

So I say it again : no one will tell you that a 24 bit DAC can resolve to 24 bits, and the unlinearity of the D/A chips is blamed.

But is it ?

Look again at the second picture.
The fun ? this is a 20 bit DAC (btw, I suspect the Metrums are). This too shows 120dBFS attenuation, but the last bit keeps on toggling (goes from 1 to 0 at the interval of the same 1000Hz signal). And to keep in mind : a 20bit DAC per se can not resolve more than 120dB (dynamic range can not be larger). So ...

I don't think this has been shown before anywhere, but while the 24 bit DAC won't resolve more because the signal will be burried in the noise, it is clear that there's still a nice sine visible, which is only made aware when it is compared to the 20bit DAC which outputs a square there (forget the peaks you see !). Aha ...

And so, at loud levels (think about an amplifier with a gain of 80(dB) or so), these both signals compare as :
24 bit DAC will exhibit a sine with some distortion which will be fairly OK;
20 bit DAC will exhibit a nasty tone because a square contains all frequencies in the world.

All right. I went fairly much off topic, but it should show you that here too noise is important and mind you, the few uV of noise I show here really (really !) is not much. But still it is important. For my second subject of this post theoretically, but for jitter vastly.

Bob, and also Jud, I will come back at your questions later. First some music for now ...

Peter

[AlainGr]

I am all eyes too !

Alain

[PeterSt]

But in brief it comes down to taking out the noise before it can harm and next have the clocks (oscillators) positioned so that no other means of jitter can be introduced.

Does this also keep noise off of ground (at least from being transmitted through or getting into the DAC via ground)?  But then one would have to provide separately for noise coming from the mains into other parts of the system, I assume.

This is more complicated. Think like the "noise" we talk about are bursts of current while current will only flow when something "draws" on it (no user no current). Also, current can only flow when the electrical circle is closed.
So, supposed the DAC is the user implying the current, then before the noise reaches critical parts the circle must be closed back to the source (read : PC).

Do we now understand ground loops ?
They are *useful* !

Peter

[PeterSt]

Your comments are really helping - before I believed that your controls changed the sequence of 01's.

Haha. No way. Well, it can with changing the volume as an example, but in 100% of cases all what I do is "lossless". This means that any processing applied - that virtually resulting in a new file to play - can restore the original from there. This is key to good sound because all remains as consistent as it was.

So, hopefully, so I can understand the computer noise / jitter issue without being a big pain in the butt, would we theoretically eliminate the issue if we had a separate device / hard drive that recorded the bit-perfect but time flawed data from the computer and that theoretical separate device could theoretically replay the data without computer noise / jitter ?.

Yes and no. Yes because when the PC is not there it wouldn't be able to produce noise - No because what newly needs to happen inside of the DAC now again creates the noise.

There's a small example of this on the forum, and it is about the galvanic isolation between PC and DAC through glass. Well, wasn't that your subject. So, the sound got way worse from it. How ? just because the the converter at the DAC end (a smallish thingy) creates more noise than the inherent solution for the NOS1 already was to get rid of the electrical noise through USB.
So yes, I already mentioned this example, but it is not theory only and sort of testifies what will happen with your solution.
Btw, in this case (an Adnaco device with USB glass connection) I predicted 100% in advance what would happen (ask Nick or Paul or AlainGr from this topic), so I guess my estimation what will happen with your solution is the same (won't help much).

And then still it is more complicated because in my (quite persistent) view, because adding noise can be for the better. Not generally, but once it covers other nastyness. This is how my plot from yesterday (the red one) serves as an example of more noise being better, so the square won't exhibit. This, opposed to less noise being better for the white plot because that nice sine will be rendered better.

All is not so black and white ...

PS - During my research I came across another "ah-ha" moment of another reason why CD seemed to sound worse over the years with processing (especially during loud passages) when it was explained that the processing interpolations could at times extend past the dynamic range of the system and cause "digital clipping".  Is that what peak extension is for?

Mwah, you are sort of correct, but then first that "interpolator" has to be wrong. Well, (as usual ) sort of, because any self-respecting interpolator would not make math errors (it's just that) in exceeding the digital domain in question (like it is not to exceed 32767-decimal for 16 bit data). But something else is going on;

The data in the music file can already be wrong for various reasons, all leading to this clipping you talk about. This is more hard to explain, but in any even it has to be the playback software correcting it. And now it becomes dangerous, because the firmware in a CD Player or DAC just has to do similar. When not, it's TICK times.
What more easy would it be, though, that the interpolator chokes on very high transient music (think synths). I really have the examples of beautifully sounding synth music which sounds completely distorted on a normal (uhm 20K++) DAC. Fun is : when you didn't hear it through the NOS1 you could easily think the distortion was normal synth sh*t. Anyway, this is not clipping, but the "reconstructor" going wild.

Peak Extension serves another purpose and although active for HDCD (within which Peak Extension is an extisting phenomenon), for general XXHighEnd usage it can serve as an overshoot protection;
Aaaaas usual there's more to say about this, and it is related to the filtering. So ... the Arc Prediction filtering does not only not ring - the effect of it is that transients nicely remain (which would otherwise smoothen into ringing in linear fashion). Now, when any not-all-so-fast DAC (mathematically) does not process any filtering because Arc Prediction is ahead of that (don't ask !), the net result will be that this DAC is confronted with higher transients than it was made for. Now, when the digital level is "full scale" regarding a transient just happening, this goes so steep against the slow DAC that it may overshoot against the voltage available. So, suppose that maxmum voltage is 2V but the sluggyness of the DAC implies an overhoot to 2.2V the voltage clips (similar to amplifier clipping). Now, attenuate somewhat (which is what Peak Extension is doing) and maximum utilized voltage will be e.g. 1.6V and a possible overhoot will be 1.8V without problem.

Mind you, nice story (bit of blah as well), but only in order with a wrongly designed DAC. Mwah, almost, because no DAC designer anticipates high transients like Arc Prediction can produce them. Just look here for the transients output once really there, and trust me that any normal reconstruction filter makes a 100% pure sine of this.

Regards,
Peter

[PeterSt]

I found two comparable picture of the same "pulse" signal, the first through Arc Prediction, the second through a common reconstruction filter.

Notice the trick here, where the pulses are not evenly spaced. Now see how that chokes the common filter. Also notice the general wave you see in the second picture (look at the peaks) and see how that is a slow frequency within itself. All not good ...

[Jud]

Those last two Dirac pulse reconstruction graphs are just really, really nice at bringing home your point.

So here's something I was thinking of when I asked the question about grounding:  Is it at all possible for noise to enter the DAC through ground on the outputs?

[PeterSt]

Ah, good question Jud.

Theoretically, why not. But practically I wonder. I think the impedance difference between the output of the DAC and the input of the amplifier (or preamplifier) is (or should be) so high that any noise induced by the amp(/preamp) will nicely circle between the output and amp(/preamp). But I am far from sure and it will be hard to measure practically.

Regards,
Peter

[bdohmeyer]

Peter,

Many thanks. I'm pretty sure I understand your computer noise answer illustration.  Your "yes" in theory was reassuring.  I also like the idea of being able to adjust the noise / jitter with the controls.

On the digital clipping, your approach / answer is similar to the explanations of, say, 2 samples that are already at a peak output where the interpolation puts a curve upward and above between the two points (I think they called it inter-sample peaks).  Two of the DAC makers that I researched say this can be corrected for by lowering the overall predictions so that the interpolated curve above the two already-peak samples is still not above the theoretical peak (they acknowledge that it cannot correct for the ADC side).

Your explanation of peak extension sounds similar.

This is fascinating - thanks again for your genius and patience

Bob

Edit: If the 2 samples are already peak - wouldn't a better algorithm draw a more straight line instead of an upward curve?