USB audio cracked... finally!

[acg]

1 hour ago, jventer said:

I think it is fantastic that people from all over the world can share on this forum.

I have been fortunate to meet Anthony (acg). I have also been fortunate to listen to his PC and his Phasure dac. We had a GTG and I heard this with a Gryphon amp on Lenehan speakers and I can still recall the details! (no loss of audio memory there!)

Thanks for sharing and please keep posting, the Phasure is special. I have no doubt that the Lush will be a hit.

 

Hi Johannes,

 

Long time no see.  The NOS1 you heard was two upgrades ago...it sounds truely amazing these days and you probably would hardly recognise it.  I'm sure we will catch up again soon, maybe at Mikes if he ever hosts another event (or you might like to visit once my horns are finished).

 

Cheers,

 

Anthony

[mmerrill99]

4 hours ago, PeterSt said:

In order to not make this 100% ballony, think oscillator output. This can be square or sine, and both will work for almost all applications. Now sit back and try to reason what will work better for your application.

Yes, the raw output from crystal oscillation is a sine wave which has to be squared up for square wave output.

Why is it changed to a square wave? Because digital signals operate on the principle of crossing detection & the moment of crossing is the timing element in all of this. Hence the slope of the rising & falling edge of square waves is far faster than sine waves & therefore should be far easier to be precise in the crossing timing. But & it's a big but - a small change in the square wave waveform will effect the timing aspect more than a small change in sinwave waveform.

 

So, we are left with considering which is more crucial (& I suspect this is what PeterST has investigated) - a precise timing of crossings but one which is upset by small changes Vs a less precise timing of crossing which is not so upset by small changes. One could call it "timing dither" - you heard the phrase here first  

[acg]

19 minutes ago, mmerrill99 said:

Can you detail your thinking on this?

Do you mean the software side or the hardware side?

 

Pretty sure Peter means software...but I may be barking up the wrong tree...he has mentioned a few tidbits in that (software) regard over the years.

 

Why do different operating systems and even versions of operating systems sound different?  Software.  Same hardware but different (operating system) software.

[pkane2001]

11 minutes ago, marce said:

Nope, only by altering the bit pattern using DSP or similar.

Well, that's what I think. But there are claims being made about USB cables, USB isolation and regen devices affecting SQ. While I've not found that to be the case, I'm always willing to try to test if there is even a small chance that it might work  Even if I prove nothing, I'll at least learn something in the process.

 

[mmerrill99]

13 minutes ago, marce said:

Rubbish, its a differential signal its the differential impedance that is critical.

I'm afraid you are wrong. We could get into a heavy discussion of how differential signalling works using schematics but it would leave the readers comotosed - instead I would refer you to the following paper with schematics of LVDS  differential circuit operation which is the same for USB operation except impedances are slightly different

Quote

In its most basic form, a differential pair is two transmission lines that have equal and opposite polarity signals traveling on them.  The property that these two signals have in common is that they are equal and opposite and they are tightly timed to each other.  Beyond these two characteristics there are no other properties that matter when a design uses differential pairs.  Maintaining the equal and opposite amplitude and timing relationship is the guiding concept when using differential pairs

Quote

 those who don’t understand how this circuit works mistakenly conclude that a 100-ohm differential impedance is required, when, in fact, what is needed is two 50-ohm transmission lines each terminated in 50 ohms

 

[mmerrill99]

15 minutes ago, pkane2001 said:

Well, that's what I think. But there are claims being made about USB cables, USB isolation and regen devices affecting SQ. While I've not found that to be the case, I'm always willing to try to test if there is even a small chance that it might work  Even if I prove nothing, I'll at least learn something in the process.

 

One thing I already pointed out to you with your ESS based DAC - the default operation for the ESS DAC chips is to engage their on-board ASRC & this will mask jitter changes in the input signal so not a good platform for testing jitter effects on the analogue output of DACs (if jitter changes are being searched for) 

[pkane2001]

4 hours ago, PeterSt said:

 

Hey hey ...

 

I think it was you who responded with a "Cool!" from a conclusion you made yourself; it was a bit hard for me to reply to that because of the way you put it there (as I recall it). But now I can do better. 

 

Low-pass filtering is not going to work out because you will eliminate the data itself. So say that USB transmits 480mbs then you can see this as equivalent to 480MHz in order for the bits (which is analogue square wave) to pass through the cable. And, as someone else nicely presented : plus a bit more or else the square wave is a sine.

So low pass filter that to e.g. 500KHz and no data arrives in the receiver.

 

IIRC you responded to my "tunnel" presentation, which implied "shaping" as such. Call it clipping and technically (electrically) you are where I like to to be. Mind you, this is not at all the same as "band limiting" which is what got to your mind, if I understand correctly.

But no harm done, because the cable already exists.

 

In order to not make this 100% ballony, think oscillator output. This can be square or sine, and both will work for almost all applications. Now sit back and try to reason what will work better for your application.

My promise : now there is suddenly so much involved that a group of fine engineers will get mad from all the ideas each one may bring to the table.

 

 

Hi Peter, of course. Band limiting below 480MHz will destroy signal and we do want the bits to get to the other end (mostly) unharmed

 

[mmerrill99]

2 hours ago, acg said:

Pretty sure Peter means software...but I may be barking up the wrong tree...he has mentioned a few tidbits in that (software) regard over the years.

Yea but they become intertwined at a sufficiently granular level - look at the USB high speed protocol (software) - a microframe every 125micro seconds giving rise to current pulses at 8KHz which is the electrical signature seen on the output of some USB audio devices

[marce]

1 hour ago, mmerrill99 said:

I'm afraid you are wrong. We could get into a heavy discussion of how differential signalling works using schematics but it would leave the readers comotosed - instead I would refer you to the following paper with schematics of LVDS  differential circuit operation which is the same for USB operation except impedances are slightly different

 

90 ohm differential impedance does not equate to 45 ohm single ended... sorry just dosen't work out that way. I have Eric Bogatins books, Howard Johnsons books, even the full tome by Lee Ritchley, point me to where it says that for differential signals it is more important to get the impedance down to the return path rather than the differential impedance. Back up with specific facts not just a whole book... That you presumed the single ended impedance was half the differential impedance shows you have a lack of understanding on diff signals. USB (the same interface works for everybody and the same specs, audio is NO different whatever some may say) so you don't need a different spec for audio, the spec is to do with the digital signal transmission... Now when a diff pair is transmitted there is no need for a Ground Return path, the pair will use each other for return signals, Ethernet over CAT5 etc.

I also am aware that only USB data transmission is diff pair, the handshaking is single ended hence the ground line is required.

[Panelhead]

8 hours ago, PeterSt said:

 

FWIW, I don't think so. It is the protocol doing the damage (that's a bit tough but alas) and that won't change a thing.

Btw there won't be much difference with using a PCIe interface with USB outputs. And on that part, we (over at Phasure) tried many things regarding this, up to everybody using a Silverstone USB3 card, just because it sounded special. But other elements caught up with it and I don't think many use it any more.

The thing here is, once you can hear a bit more fro your system, everything matters regarding USB. Buy 10 different PCIe cards and they will all sound different and you will keep one as your best one. Until the Operating System changes - then you have to start all over.

Buy a JCat Femto etc. card - that too sounds different.

 

In the end those "tweaks" don't work, because they work by random aids. Your system suffers from something, and the aid will change the sound. Will it be for the better ? actually you will know when you e.g. buy another DAC and suddenly it does not work for the better any more.

From the point of view of the manufacturer it can be looked at differently with the same conclusions : if something works for really everyone, then the application is really good in absolute sense. And the fun is that when working in a quite close group, this is easy to detect (but also carries great responsibility). What I mean is, once an application has been worked out and it works for a 100, then when a first comes along where it does not work, then

a. I need dare to say that something has to be wrong at the customer's end;

b. The customers needs to have an open mind and help searching for the culprit.

 

A bit of a blabla story but this is really how it can work and how everyone can help each other and make progress.

Back to the quote above, what can be predicted is that the speed of e.g. USB3.2 helps to overcome problems in general (the speed is by far not needed for audio but spades of headroom is always a good thing). This is (in my view) how USB3 always sounded better than USB2 (at the PC's end) and how a USB3 hub in the ISO REGEN sounds better than a USB2 hub in the same environment. We shouldn't make this a subject in this topic, but a lot with audio (if not all) is about headroom we don't anticipate because we can so-called calculate it isn't necessary.

  The use of USB-3 may not be very helpful. Just the increased data rate capability should offer advantages. 

  USB-C transmission is different. This can be Thunderbolt 3. This can offer benefits. The 40 GHZ is one, the PCI-e  protocol is another. 

  But this is topic for another thread.

  

[jabbr]

1 hour ago, marce said:

90 ohm differential impedance does not equate to 45 ohm single ended... sorry just dosen't work out that way. I have Eric Bogatins books, Howard Johnsons books, even the full tome by Lee Ritchley

Yes. as you say, detailed in Howard Johnson's "Black Magic" and "Advanced Black Magic" books on high speed digital design  There are two issues:

1) differential pair impedance

2) termination  impedance -- (can be to ground)

 

https:\\www.micron.com\~\media\documents\products\technical-note\dram\tn4606_point_to_point-termination.pdf

Note Figure 20, as the rise time of the transition slows the SI improves ... this is an example of removing "hard edges".

 

https://fenix.tecnico.ulisboa.pt/downloadFile/3779579943588/[Bus Terminations] Ethirajan and Nemec - Termination Techniques for High Speed Buses.pdf -- see Figure 4 for use of capacitance along with resistance.

 

http://www.ti.com/lit/an/snla043/snla043.pdf gives an extensive discussion of the effects of and use of termination capacitance.

 

I wouldn't at all be surprised if audio designers are not using high speed digital design techniques -- who knows what termination is being used, but I wouldn't be surprised if at best resistive and as demonstrated in the Micron paper I referenced above, it is very possible that this has deleterious effects on the USB SI at the receiver.

 

I have no idea how the "Lush" actually works, but it might include its own termination, or as also referenced in the above papers, the transmission line properties may have similar effects.

 

@PeterSt -- the above papers detail what I was saying regarding capacitive termination -- all of this would need to take into account the cable impedance (and transmission line model) itself.

[mmerrill99]

@marce - did you read Lee Ritchey's article which I linked to & from which I directly quoted?

You seem to be saying exactly the opposite of him & stating that he has a lack of understanding as that was who I quoted in 100ohm differential vs 50ohm single ended impdance. You appear to be making the exact mistakes that he states engineers usually do

Quote

those who don’t understand how this circuit works mistakenly conclude that a 100-ohm differential impedance is required, when, in fact, what is needed is two 50-ohm transmission lines each terminated in 50 ohms

 

But seeing as you are averse to what he says in that article (or didn't read it), maybe this one, again from Lee Ritchie, will more directly address your issues "Differential Signaling Doesn’t Require Differential Impedance"

 

In it he states:

Quote

As it turns out, differential impedance doesn’t play a role in this form of signaling and is not necessary. How it was decided to impose this condition on this type of signaling is a mystery. It probably happened because the word differential is in the title. In any case, as will be seen in the following explanation of how this signaling works, other design considerations are more important.

 

[elcorso]

6 hours ago, manisandher said:

 

FWIW, I have a strong preference for using the USB2 port at the back of the audio PC - it sounds more 'alive' than any of the USB3 slots, with a bit more bite. USB3 sounds full-bodied, but overly-smooth and boring in comparison... in my system... to my ears. (And yes, even with the Lush feeding the Phisolator built into the DAC, different PC ports still sound different.)

 

Mani.

 

Like SE vs BAL in analogue?

 

Gentlemen, please ignore my post, I don't want to derail this thread !

 

Roch

[mmerrill99]

Maybe a quote from Altium's pcb routing software might convince?

Quote

According to Lee Ritchey, a noted industry high-speed PCB design expert, successful differential signaling does not require working to a specific differential impedance. What it does require is:

• To set each of the routing signal impedances to half the incoming differential cable impedance.
• That each of the two signal lines is properly terminated in its own characteristic impedance at the receiver end.
• That the two lines should be of equal length, to within tolerances of the logic family and the circuit frequency used in the design. The focus should be on preserving the timing, match the lengths close enough to satisfy the skew budget of the design. Example length tolerances include: high-speed USB, length mis-match should be no greater than 150 mils; DDR2 clocks need to be matched to within 25 mils.
• Use the benefit of routing the two signals side-by-side to help achieve good quality routing of matched lengths, where required it is acceptable to separate to route around obstacles.
• Layer changes are acceptable, as long as the signal impedances are maintained.

 

[jabbr]

5 minutes ago, mmerrill99 said:

But seeing as you are averse to what he says in that article (or didn't read it), maybe this one, again from Lee Ritchie, will more directly address your issues "Differential Signaling Doesn’t Require Differential Impedance"

 

 

Single ended impedance on a differential pair requires a common ground between transmitter and receiver  

[mmerrill99]

2 minutes ago, jabbr said:

 

Single ended impedance on a differential pair requires a common ground between transmitter and receiver  

As far as I can see there is common ground between transmitter & receiver ala the grounding of the circuit, not necessarily a specific ground wire.

 

Did you read the Lee Ricthie articles I linked to & looked at the schematics in them?

 

That's why I linked to them so that these questions could be seen to be answered in the article without bringing such an esoteric topic here which will bore the pants off most  readers!

[mmerrill99]

32 minutes ago, jabbr said:

I wouldn't at all be surprised if audio designers are not using high speed digital design techniques -- who knows what termination is being used, but I wouldn't be surprised if at best resistive and as demonstrated in the Micron paper I referenced above, it is very possible that this has deleterious effects on the USB SI at the receiver.

No audio designer I know designs the USB receiver itself - it is a bought in chip which has correct termination resistors already built-in

[PeterSt]

4 hours ago, mmerrill99 said:

Can you detail your thinking on this?

Do you mean the software side or the hardware side?

 

Hi mmerrill,

 

I don't think I can add much to what is already known;

 

It is about the 8KHz which is right in the middle of the audible spectrum and thus can never be filtered out. This in itself creates the plethora of solutions to counter attack that noise (which really sticks out its neck when normal system noise drops under  ~-140dB.

 

It is about the communication itself which has its own life. So the packets and how much data they contain and thus also how much current is drawn per time unit; this always changes and changes.

 

It is about (similar as above) the unconnected clock speeds at both ends with which all needs to deal with (this is possibly less known).

 

It is in the end about the somewhat unnecessary difficult protocol in itself. I mean, no audio sending data device requires it and no audio receiving device requires it. Someone just thought that USB was a good idea.

Of course it was at first because it eliminated the soundcard as such (5 million people may not even understand the importance of this), but the sound had also to be good.

And so we actually still fight a bit with that.

 

[jabbr]

17 minutes ago, mmerrill99 said:

As far as I can see there is common ground between transmitter & receiver ala the grounding of the circuit, not necessarily a specific ground wire.

 

Did you read the Lee Ricthie articles I linked to & looked at the schematics in them?

 

That's why I linked to them so that these questions could be seen to be answered in the article without bringing such an esoteric topic here which will bore the pants off most  readers!

He is discussing PCB and agreed that it is often desirable not to route differential signals together e.g the + and - of a balanced signal have go to different physical areas to be processed.

 

The 1/2 impedance desires where a "twinlead" transmission line e.g. Ethernet meets a PCB. Each line is terminated and then go on to PCB traces. You could have twin coax but that's not how Ethernet (nor USB I presume) is specified. The spec is differential impedance 100 (and 90) Ohm.

[PeterSt]

4 hours ago, mmerrill99 said:

Not sure this holds up - in USB 3 receiver chips there seem to be two separate engines - a USB 2 engine & USB 3 engine - they each handle their own protocol independently of the other. USB 2 only operates at 480MHz, no faster

 

Your more general point that the chip itself is built to be capable of handling the faster speed USB 3 (5Gbps?) may have some significance though even when it's only handling USB 2.0  480Mbps speed throughput?

 

Faster settling time of on-chip components could well be significant here?

 

All matching what was on my mind when responding.

It goes a bit beyond this when you think of things like the heat-up which is now insignificant (and thus also does not change properties once playing) and what about jitter specs which again also matter.

Regarding the latter : ... *if* jitter matters indeed (and very indirectly it will but this is far sought).

Notice that with jitter I again refer to the propagation times (which are just 10 times faster), and the skew on the two data lines which now also has a lesser distance (and this results in jitter on the net data line).

 

Anyway, you can see how over-sizing almost always helps with audio, and also how these things are nog easy to guess for consumers.

[marce]

10 minutes ago, mmerrill99 said:

@marce - did you read Lee Ritchey's article which I linked to & from which I directly quoted?

You seem to be saying exactly the opposite of him & stating that he has a lack of understanding as that was who I quoted in 100ohm differential vs 50ohm single ended impdance. You appear to be making the exact mistakes that he states engineers usually do

 

But seeing as you are averse to what he says in that article (or didn't read it), maybe this one, again from Lee Ritchie, will more directly address your issues "Differential Signaling Doesn’t Require Differential Impedance"

 

In it he states:

 

Yes I have read it, its more to give an overall better signal if you can't keep the lines tightly coupled. And no download Saturn PCB toolkit and work out the impedance's (or the Polar tools),differential pair structure does not give single ended impedance half that of a the diff pairs impedance, this is where I misread your comment, to get two 50R lines requires different geometry than a 100r pair. . (Any date on the second paper ECL logic...) yes I am not adverse to  Lee Ritchleys stuff, but tend more towards Howard Johnson and Eric Bogatin.

Why do all the differential interfaces demand a set differential impedance, and try routing the signals as signle ended and see what the EE in charge will do, we may be operating delusionally doing diff pair routing on a PCB, but it keeps the whole impedance seen by the signal as equal as possible over PCB, cable etc. and we know it works, its been well documented and most follow the quite strict guidelines for characteristic impedance, skew etc.As said these guidelines are for PCB traces, where coupling is not as effective as it is with twisted pair cable, where high percentage coupling gives the best common mode noise rejection.

Jabbr, we use SIV software to determine if any signals have problems and whether termination is needed, USB1 & 2 is pretty robust, relatively low speed, USB 3 brings more problems as it has to run at higher speeds, not always an advantage.

 

[mmerrill99]

4 minutes ago, jabbr said:

He is discussing PCB and agreed that it is often desirable not to route differential signals together e.g the + and - of a balanced signal have go to different physical areas to be processed.

You mustn't have read this section then?

"

Quote

 

Handling the Transition from Twisted Pairs to PCB Traces

Much confusion centers around how to handle the transition from twisted pair transmission lines to transmission lines on a PCB. Figure 3 shows several types of transmission lines. In all four cases the electromagnetic field travels between the two conductors that make up the transmission line. All four types are characterized by specifying the impedance measured between the two conductors.

 

 

[marce]

The lines of differential or balanced signal should be run together, in cable they should be twisted pair. As the various papers show, PCB design is often a compromise, but most run differential signals as closely coupled pairs. With cables and the close coupling each cable provides the return path for its closely coupled mate.

[mmerrill99]

5 minutes ago, marce said:

As said these guidelines are for PCB traces, where coupling is not as effective as it is with twisted pair cable, where high percentage coupling gives the best common mode noise rejection.

As I said to Jabbr, I would suggest you read the links & correct your knowledge on this - you seem to have many contradictory views to what Ritchie says.

 

No, it's not just about PCB traces as I pointed out to jabbr

[PeterSt]

51 minutes ago, jabbr said:

@PeterSt -- the above papers detail what I was saying regarding capacitive termination -- all of this would need to take into account the cable impedance (and transmission line model) itself.

 

@jabbr,

 

Prize of the day (OK, yesterday) goes to you. Somehow you must be in the wrong job. 

For others, Jonathan sent me a PM suggesting this, which I rejected. But why ?

 

I can't judge that type of solution "brain wise" because too much of the same is involved elsewhere (maybe it is better to think "everywhere"). It's a resistor here, a capacitor there and a ferrite bead such. But not in the cable (promise).

So doing it like that would be quite out of control. But also and honestly, I just would not attempt it like that because too much of real filtering would be involved and it would destroy the signal too much (maybe).