USB audio cracked... finally!

[marce]

Simultaneous switching noise.

Digital Signal overshoot and ringing.

[marce]

2 minutes ago, mansr said:

Is this supposed overshoot real? I don't see it in any of the scope screenshots usually trotted out when promoting USB decrapification widgets.

Without measurements we will never know, that's my point about digital signal transmission, you need to look at the signal to see what is going on...

[mmerrill99]

9 minutes ago, mansr said:

Is this supposed overshoot real? I don't see it in any of the scope screenshots usually trotted out when promoting USB decrapification widgets.

"Supposed overshoot real"?

Have you looked at any of Jabbr's linked documents?

Why would you think that USB signals are not subject to the same considerations as every other high speed digital signal?

 

Edit: Remember what we are discussing are possible mechanisms so looking for evidence now is contrary to the theoretical nature of the discussion

[PeterSt]

Just now, PeterSt said:

Also keep in mind, that in my first posts in this thread I have been repeatedly (kind of) claiming that I regard jitter in USB about completely irrelevant. But still a little about that in a next post.

 

"about irrelevant" because you now should be able to reflect with the noise coming from the packets, as I just explained. So, you now know that bursts of packets imply current draw, at irregular interval and therefore the noise of it becoming deterministic (expensive word for making the frequencies of it visible). OK ...

 

Now let's think the exact other way around, and the stream of current spikes needed would be nicely even and although there could be a frequency of it visible, it is one frequency only, plus it is of low level because no "bursting" is in order and the current requirement is spread over all the bits (and lets assume it is all 1-0-1-0-1-0-1-0-1-0 etc.).

Next up is jitter in that digital data ...

 

The 0's and 1's are still nicely interpreted and no bits flip. However :

The trigger point on the slopes of that signal is now higher and lower *because* it is jitter doing that (this may require some choking thinking, so think about it). Meanwhile, because the trigger is at the e.g. higher point of the slope, the current force required to get the slope where it is, has already been accomplished. This is how what's triggered itself, starts to prevail on the current draw.

So what I say here is that to get the slope all the way up right from the base might require 5mA and at the normal trigger point is still requires 1mA to proceed to its peak. With jitter, and the slope being a bit higher already, it requires 0.9mA only (so if we see it as a constant draw, it is 0.9mA at that point).

Now comes our trigger, and whatever it is what is triggered, it requires 0.2mA ...

 

It should be obvious in this example that when the constant current draw is 1mA that 0.2mA is 20% of that;

when the constant draw is 0.9mA the 0.2mA is 22.2%.

 

So in this example the impact on the current of what is triggered, varies 4.4% because of jitter.

And now we have the deterministic noise again.

And the rest.

 

I hope you can see that nobody in the world will be able to really work with this, because it is a complex which is, well, too complex. With the last example I assumed 1-0-1-0-1-0-1-0 etc., which already is not so and it largely will debunk my example because of it. Still my story will be true and it all depends on oscillating patterns.

And patterns are the things we hear as flavors.

 

[PeterSt]

11 minutes ago, mansr said:

Is this supposed overshoot real?

 

Wouldn't it be so that any wire (or other bottleneck) which is too bandwidth limited for what the signal wants to accomplish (like infinite rise time or something - haha) ... will exhibit overshoot (and ring) ?

[mmerrill99]

8 minutes ago, PeterSt said:

Still my story will be true and it all depends on oscillating patterns.

And patterns are the things we hear as flavors.

Yep, this is what I suspect too - it's the signal correlated noise that is noticeable

How we perceive it is a whole other discussion!

[pkane2001]

21 minutes ago, PeterSt said:

The trigger point on the slopes of that signal is now higher and lower *because* it is jitter doing that (this may require some choking thinking, so think about it). Meanwhile, because the trigger is at the e.g. higher point of the slope, the current force required to get the slope where it is, has already been accomplished. This is how what's triggered itself, starts to prevail on the current draw.

 

Thanks, Peter. I appreciate you trying to provide an explanation. But, isn't trigger point driven by the voltage level? Wouldn't it trigger at the same level, regardless of slope? The only difference due to jitter is the timing of this trigger, not the amount of current drawn.

[marce]

52 minutes ago, mmerrill99 said:

"Supposed overshoot real"?

Have you looked at any of Jabbr's linked documents?

Why would you think that USB signals are not subject to the same considerations as every other high speed digital signal?

 

Edit: Remember what we are discussing are possible mechanisms so looking for evidence now is contrary to the theoretical nature of the discussion

Without actually taking a look at the waveforms of the system in question it is just guessing though at whether there is ringing... Then you have to determine how much extra noise a bit of ringing (easily tamed) is causing.

54 minutes ago, PeterSt said:

 

"about irrelevant" because you now should be able to reflect with the noise coming from the packets, as I just explained. So, you now know that bursts of packets imply current draw, at irregular interval and therefore the noise of it becoming deterministic (expensive word for making the frequencies of it visible). OK ...

 

Now let's think the exact other way around, and the stream of current spikes needed would be nicely even and although there could be a frequency of it visible, it is one frequency only, plus it is of low level because no "bursting" is in order and the current requirement is spread over all the bits (and lets assume it is all 1-0-1-0-1-0-1-0-1-0 etc.).

Next up is jitter in that digital data ...

 

The 0's and 1's are still nicely interpreted and no bits flip. However :

The trigger point on the slopes of that signal is now higher and lower *because* it is jitter doing that (this may require some choking thinking, so think about it). Meanwhile, because the trigger is at the e.g. higher point of the slope, the current force required to get the slope where it is, has already been accomplished. This is how what's triggered itself, starts to prevail on the current draw.

So what I say here is that to get the slope all the way up right from the base might require 5mA and at the normal trigger point is still requires 1mA to proceed to its peak. With jitter, and the slope being a bit higher already, it requires 0.9mA only (so if we see it as a constant draw, it is 0.9mA at that point).

Now comes our trigger, and whatever it is what is triggered, it requires 0.2mA ...

 

It should be obvious in this example that when the constant current draw is 1mA that 0.2mA is 20% of that;

when the constant draw is 0.9mA the 0.2mA is 22.2%.

 

So in this example the impact on the current of what is triggered, varies 4.4% because of jitter.

And now we have the deterministic noise again.

And the rest.

 

I hope you can see that nobody in the world will be able to really work with this, because it is a complex which is, well, too complex. With the last example I assumed 1-0-1-0-1-0-1-0 etc., which already is not so and it largely will debunk my example because of it. Still my story will be true and it all depends on oscillating patterns.

And patterns are the things we hear as flavors.

 

I don't get this...

[PeterSt]

20 minutes ago, pkane2001 said:

Thanks, Peter. I appreciate you trying to provide an explanation. But, isn't trigger point driven by the voltage level? Wouldn't it trigger at the same level, regardless of slope? The only difference due to jitter is the timing of this trigger, not the amount of current drawn.

 

You are 100% correct. I got myself mixed up with a slightly different story about the rise time, but  tried to avoid that.

Not sure how to correct this so all doesn't get confusing even more. So wait a bit ...

[mmerrill99]

3 minutes ago, marce said:

Without actually taking a look at the waveforms of the system in question it is just guessing though at whether there is ringing... Then you have to determine how much extra noise a bit of ringing (easily tamed) is causing.

As said this is a theoretical discussion of the possible mechanisms that PeterSt's cable is using.

But you should already have encountered ringing, overshoot & underdamping in your work. Have you done any such measurements on USB 2?

 

Yes, measurements would be needed to confirm or otherwise

[PeterSt]

3 minutes ago, marce said:

Without actually taking a look at the waveforms of the system in question it is just guessing though at whether there is ringing... Then you have to determine how much extra noise a bit of ringing (easily tamed) is causing.

 

Marce, can you tell me what you would see when you try to squeeze a square wave (which is supposed to have a close infinite frequency for a close to infinitely short rise time) through a too low bandwidth cable ?

There is no hidden question here and *you* are the HF designer.

"Chaos" would count as an answer, but maybe there is more between black and white.

[marce]

17 minutes ago, mmerrill99 said:

As said this is a theoretical discussion of the possible mechanisms that PeterSt's cable is using.

But you should already have encountered ringing, overshoot & underdamping in your work. Have you done any such measurements on USB 2?

 

Yes, measurements would be needed to confirm or otherwise

Yes, it can vary, again it depends on what systems/boards etc are present. I would say bespoke designs tend to be better than generic PCs tablets, which tend to be the worse.

[PeterSt]

1 hour ago, pkane2001 said:

Thanks, Peter. I appreciate you trying to provide an explanation. But, isn't trigger point driven by the voltage level? Wouldn't it trigger at the same level, regardless of slope? The only difference due to jitter is the timing of this trigger, not the amount of current drawn.

 

So with apologies and ...

 

36 minutes ago, PeterSt said:

You are 100% correct. I got myself mixed up with a slightly different story about the rise time, but  tried to avoid that.

Not sure how to correct this so all doesn't get confusing even more. So wait a bit ...

 

I hope to be able to undo the damage ...

 

When the rise time is steep, there is a certain relation between the current still needed to let rise the wave to its maximum level. Of course it is so that the trigger point (could be 0.8V on a total of 2V)  has this relation to the result of the trigger itself (implying 0.2A in my earlier example). Nothing wrong there yet.

 

But since the subject is the rise time and what it can imply, it can nicely try to avoid this, but of course this doesn't work out because indeed we can not see that the trigger point will be higher for an unknown reason. It isn't (and my example went in the wrong direction there).

So, incorporating the rise time after all :

 

When the rise time now is higher than before (to talk in the realm of the earlier example), there will be more voltage needed to get it at its highest point (still no change in the story), BUT the trigger will be earlier for th rising edge and later for the falling edge. Btw, no jitter again - it is not (additionally or less) there and it is not related.

From here the story is the same, as it is more about the relation to the "continuous" current draw is changed again.

To have it more clear, now compare with the sine; the trigger point is still at the same (voltage level) but later on the rising edge and earlier on the falling edge (eye must stay sufficiently open of course) but the current draw is way less; now the trigger and what it implies has huge impact (relatively).

 

[mmerrill99]

Just for some feel for all of this - this youtube video helps to show some issues with USB 3 supersepeed signal (5GHz) in a real world Icebox raid box USB 3. This is that Oz guy EEVblog 'messing around' with a $140,000 12GHz scope to measure.

 

I have linked to the point in the video where he starts to look at the eye pattern.

There are some interesting issues shown on this - admittedly USB 3 with a not optimal probes (about $12,000 for a differential probe)

 

I would draw attention to the very tight eye (the blackish area in the scope) in the scope plots. But also look on the color grading shots & see that there are times when the slope of the signal is way out of spec - shown by green dots in this blackish eye area - as the capture time increases these dots show more & more, beginning to fill in the blackish area - signifying that there are random signal waveforms which are very far off where they should be (you will need to use full screen display to see this)

 

Look also at the difference between the transmit USB eye pattern Vs the USB receive pattern & notice the overshoot & ringing on  the receive pattern

 

There are many other aspects worth talking about that Marce might want to comment on?

 

 

[mmerrill99]

Just in case some are unaware of what an eye pattern is & what it represents, here's an explanatory video

 

[mmerrill99]

BTW, I don't believe overshoot or ringing are part of the USB specification - are they part of conformance testing?

Overshoot or ringing will not result in an eye pattern that breaches a USB 2.0 eye mask which is one of the main tests for USB conformance.

 

In fact what I see mentioned in USB 2 conformance testing is this stipulation: "The peak should be taken at the plateaus of the wider pulses to avoid inflated reading due to overshoots" So this is a direct directive to ignore overshoots (I think they are recognised & tested for in USB 3?).

 

Is this not the issue - a restricted consideration of what's needed to deliver the bits rather than a wider system perspective about possible electrical noise issues in the analogue (clocks) side of USB audio devices?

[PeterSt]

1 minute ago, mmerrill99 said:

Overshoot or ringing will not result in an eye pattern that breaches a USB 2.0 eye mask which is one of the main tests for USB conformance.

 

I don't think it is of any real relevance, but if there is high ringing, it will narrow the eye vertically and if you now also have a narrow width, then you have a problem.

So a narrow width may not be too narrow, while additional ringing may shut down the connection (so to speak).

So ringing itself should be OK just the same. Maybe there's a "just in case" in order, but I don't think Marce will agree with that. It just has to bee "good".

 

Now see how Chinese USB2 cables are twisted, sometimes. 

And thus also see that a "just in case" is in order at least somewhere, because the whole chain is not controlled by one instance. PC should be fine, USB receiver at the other end should be fine, but our cables are not under anyone's control (well, virtually).

 

And so, as usual, everything matters.

[mansr]

3 minutes ago, mmerrill99 said:

BTW, I don't believe overshoot or ringing are part of the USB specification - are they part of conformance testing?

Overshoot or ringing will not result in an eye pattern that breaches a USB 2.0 eye mask which is one of the main tests for USB conformance.

Yes, as long as the signal stays within the designated mask area, anything goes.

 

Now it seems you people are making a number of assumptions:

• There is significant overshoot and/or ringing in typical USB implementations.
• This overshoot and/or ringing has audible effects on the analogue output of the DAC.
• Peter's cables alter the waveform somehow.
• Slightly different alterations to the USB waveform have radically different audible effects.
• Said audible effects are consistent across all DAC implementations.

Not one shred of evidence has been produced to validate any of these assumptions.

[mmerrill99]

2 minutes ago, PeterSt said:

I don't think it is of any real relevance, but if there is high ringing, it will narrow the eye vertically and if you now also have a narrow width, then you have a problem.

IMO, it's undershoot that causes eye pattern to narrow vertically, not overshoot. The ringing after an overshoot usually doesn't encroach on the eye mask

[plissken]

I'm always up for a GTG where people can bring some various gear and cabling and go through some of this in a controlled manner.

[mmerrill99]

25 minutes ago, mansr said:

Yes, as long as the signal stays within the designated mask area, anything goes.

 

Now it seems you people are making a number of assumptions:

• There is significant overshoot and/or ringing in typical USB implementations.
• This overshoot and/or ringing has audible effects on the analogue output of the DAC.
• Peter's cables alter the waveform somehow.
• Slightly different alterations to the USB waveform have radically different audible effects.
• Said audible effects are consistent across all DAC implementations.

Not one shred of evidence has been produced to validate any of these assumptions.

Again, does it need stating that this is a theoretical discussion - looking for evidence is the next stage & what one does to confirm or deny proposed mechanisms, right? Citing a lack of evidence at this stage is moot & rather jumping the gun, don't you think? (Maybe PeteSt already has the measurements - I don't know?)

 

Do you have a $140,000 scope you can lend anyone to test these proposed mechanisms? 

 

So let's take each of your points:

• What do you call significant? As stated if overshoot/ringing is not part of USB conformance testing how do we know what is typical?
• It's proposed that the overshoot/ringing may itself have effects on the ground noise through any number of mechanisms but mainly via the USB receiver & the ground noise (often called ground bounce) that is produced when dealing with signal integrity issues. Have you simulated or measured what effect of overshoot/ringing has inside a USB audio device?
• What Peter's cables are doing is what the thread is about?????
• You seem to only consider linear consequences i.e that the amplitude of the input signal is directly proportional to its resultant effect? Who said it has to be a linear effect? Is jitter a linear effect?
• If overshoots/ringing are common issues in USB SI at the receiver & there are a limited number of USB receivers used in USB audio then it's reasonable to assume that there is a certain consistency across USB audio devices

[jabbr]

3 minutes ago, mmerrill99 said:

Again, does it need stating that this is a theoretical discussion - looking for evidence is the next stage & what one does to confirm or deny proposed mechanisms, right?

Strong second to that. @PeterSt is being understandably coy about discussing what the cable actually does and I suspect his "explanations" are not designed to be precise such that the IP could be copied. As such we are playing a mind reading guessing game.

 

This discussion does highlight many misconceptions about digital signaling eg rise time is specified as a minimum of 500 picoseconds

[PeterSt]

41 minutes ago, mmerrill99 said:

The ringing after an overshoot usually doesn't encroach on the eye mask

 

I had a relatively graduate slope in mind. But this is exactly where ringing (overshoot) does *not* occur.

Thanks.

[pkane2001]

28 minutes ago, mmerrill99 said:

looking for evidence is the next stage & what one does to confirm or deny proposed mechanisms, right? Citing a lack of evidence at this stage is moot & rather jumping the gun, don't you think? 

 

I'm all ears: how do you propose we go about confirming or denying the proposed mechanisms?

[mmerrill99]

16 minutes ago, pkane2001 said:

 

I'm all ears: how do you propose we go about confirming or denying the proposed mechanisms?

Well, first a proposed mechanism of operation would be conjectured & then a suitable means of measuring such would need to be proposed.

 

So first off would be to characterise the USB signal at a number of USB audio receivers using a number of USB cables to check for overshoot/ringing.

The next step would depend on the results of this.

 

But if you really were "all ears" then a much simpler test would be to acquire one of the LUSH cables & listen

Bring it to as many different USB audio installations as you have access to & listen

 

This is much better empirical evidence than the measurements which will be argued over - (I've no doubt that the listening results would be argued over too)