Ethernet vs USB

[jabbr]

You may get the idea that I'm a fan of keeping my upsampling workstation away from the electronics using an Ethernet 

Haha... I remember trying to convince Peter to go the NAA route some 4 years ago:

So where does this leave us? Here’s my ultimate dream setup:

- main PC with W7/W8 (shouldn’t matter at all)
- although HQPlayer is cable of excellent SQ, and even though most of the controls in the Settings area should become obsolete, I’d rather use XX
     - I prefer the library management
     - I really, really want to use phase alignment
- of course, the NOS1 DAC – it’s simply the best I’ve heard

What would it take to achieve this dream setup? Obviously, Peter and Miska (Jussi) need to work together on it, and find a mutually beneficial way to do this. There would need to be appropriate drivers made for the NOS1 to work under Linux. And I think that’s it...

But a lot has changed since then. With the Mach II audio PC, I'm not sure an NAA is necessary now. And in any event, without a Linux driver, we can't use our NOS1s with an NAA.

But USB remains to total PITA, and if ethernet is the ultimate solution, we should pursue it. But audio PC -> ethernet -> NOS1 is surely the way to go (as indeed you seem to suggest).

Mani.

You are on to something. The "NAA" is a small program that implements an audio driver remoting protocol that accepts Ethernet packets, buffers them, and writes them to the audio driver. It also queries the driver to make it appear as if it were locally  attached to HQPlayer ... in the case of XXHE or any other audio program for that matter, a remoting layer could appear as an audio device driver and send the packets across the network as opposed to the USB cable. The difference is that fiberoptic Ethernet could travel kms without concern... and if 1 Gbe isn't fast enough there's 100Gbe and higher.
So Ravenna does all this but at significant cost. We could do the same with regular Ethernet & software.
The endpoint could run Windows but there are chips that combine ARM & FPGA (Zynq) and so everything can be done with one part and that reduces development cost dramatically (there is still the programming)

Now Miska ideally would cooperate on an open "NAA" protocol or maybe he would give Peter the specs, or else a similar protocol could be developed.

Jonathan

[manisandher]

The endpoint could run...

And this is the issue for me. Whether the endpoint runs Windows (great, we could still use our NOS1s) or Linux (a NOS1 driver would have to be developed), there would still need to be a connection from the endpoint to the DAC. Unless, of course, the endpoint is built into the DAC, which might introduce all sorts of noise issues into the DAC itself.

To me, it's absolutely not clear why ethernet, I2s, spdif, AES/EBU, or any other audio connection would deal with noise inherently better than a USB connection. The galvanic isolation of some of these says nothing about their ability to reject noise, and it's noise getting through to the DAC chips that can be the only mechanism responsible for the changes we hear in sound when we make changes to software/hardware upstream.

My gut feeling is that we (Peter and others!) have taken USB so far now that it might actually out-perform all other audio connections right now. Just a gut feeling.

Mani.

[briefremarks]

Quoting something from a reddit forum (FWIW)  from Antipodes on design of their music server:

"In the context of Antipodes music servers, USB is the best connection. With a noisy server (most other servers), using an Ethernet input to the DAC makes sense because it provides the greatest isolation from noise interference with the digital signal. But Ethernet generates high levels of noise within the DAC. This means that Ethernet is a band-aid, not an optimal solution.

SPDIF, i2s and AES/EBU suffer because the system clock is in the server, not in the DAC where it really should be, and this is typically heard as dirty and less coherent sound.

USB is ideal for an Antipodes server because it is architecturally superior to SPDIF, i2s and AES/EBU, and generates much less noise in the DAC than Ethernet. But it needs to be emphasised that USB is the best solution only when the server is very low noise, and when the DAC manufacturer has done a competent job of isolating the USB receiver from the analog power supply and circuitry - fortunately most have. But some DACs may sound better via the SPDIF digital output of an Antipodes music server, for this reason. Using Ethernet will always sound softened with images being less distinct."

[PeterSt]

Hey Mani,

Maybe for some it goes unnoticed, but you say/claim maybe more important things than visible at first glance. And again by sheer coincidence I have the proof at hand - just from last night.

I tried another self-made USB cable. Say with different topology and just off-spec. It worked error-free. This latter should mean : digital = digital so what can go wrong.

Well, you'd drop dead if you had heard it.

Already at the first hit of a cymbal it was super clear that there was not only more metal but also more colour in the metal. Usually this doesn't come both at the same time (more metal takes out colour).
There was more bass which usually is an indication that something is not right with USB (100% of people think that more bass will be for the better, except me).
By high exception I listened to it while working, just because I was excited for the outcome of the cable. After 30 minutes I had to shut off the music because something was too much hammering on my mind. Too much energy or something.

In the evening's normal listening session I had to say to myself "no Peter, this time you will let the cable in instead of taking it out within 5 minutes". So I did.
After an hour I head the idea I got used to it and that for some music it worked out for the better. But I really had to be careful with the music selection. For example, Year of the Cat just was ugly, while this album normally carries beauty all over. W8 robots playing came to my mind once again.
In some very strange fashion there was too much energy, up to standing waves in bass as well as higher frequencies (all buzzes - very very similar to using another player with too much ringing).

Meanwhile I thought of the name for the cable and came up with a simple "The Stereo". And *that* was the attracting thing : the stereo image was ridiculous. So this time no ridiculous hall, but "way much stereo". Man, how to bring this across.

After 3 hours or so, dinner almost ready, the person at the bar dropped off the headphones because finished with watching something else (with sound) and right away asked me what the hell I had done. But the question was moot because that same person made the cable and it needed no answer.

Anyway the message thrown was that Bose milk packs most probably would sound better.
That did it.

The one but last thing we did was adding the second Phisolator which did not help a thing and the very last thing was putting back the Clairixa.
Wow. All twinkels and bells readily back and all normal and with ease.

So there you go. I am now able to make a USB cable which is just a little bit off spec (think impedance) and the difference is so crazy that you don't know where to go.
Zero errors.
But also this : one of the characteristics of the topology is that it was not a stranded cable. And you can just hear what this does (with some common sense and experience). Maybe some recall the speaker cables from (non-stranded) coax ? (people started to make them from transformer wire, remember ?) - the sound was similar, as I recall. How something like that can be audible through a digital (USB) connection - it will be coincidence.
But I did this for a reason of course and I am sure it has potential.

Tonight I am going to listen to another cable, which will be build up just a little bit differently ...

Anyway moral : yes Mani. I think we are quite far already with what we achieved with USB. But if you see what can be achieved for enormous difference just by tweaking the USB cable, then who knows what may come of that.

Peter

[manisandher]

Anyway moral : yes Mani. I think we are quite far already with what we achieved with USB. But if you see what can be achieved for enormous difference just by tweaking the USB cable, then who knows what may come of that.

Hey Peter, good luck with this. Over the years, I've spent an inordinate amount of time comparing a whole load of different BNC-spdif cables to each other and XLR-AES cables to each other. They've all sounded different! Changing the length of the cable (i.e. exactly the same cable material) seemed to change the sound. My feeling was that these changes must be due to impedance mismatches at the connections, and that longer lengths can actually be helpful here.

FWIW, I think my longer Clarixa (3.5m) is at least as good, if not better sounding than the shorter Clarixa cables I have. Perhaps down to an impedance mismatch again??? If you've never tried a longer length, perhaps you should?

Mani.

[jabbr]

My approach to cables is that when/if cables make a difference it is a symptom of an underlying problem that should be properly fixed. So fiberoptic Ethernet "fixes" a whole category of possible problems. Now if optical cables get impedance mismatches we have bigger problems

[PeterSt]

Now if optical cables get impedance mismatches we have bigger problems

Hi Jonathan,

Yeah, you say it in a joking fashion, but I know someone in one of the states of your country who makes a living out of reflection testing of fiber cables (I have seen his gear).
I am not sure but I think fiber needs damping too (sunglasses ? haha).

Anyway, it is my kind of promise that only part of it is about the cable itself. The largest part is at the end point(s) where noise is re-generated. For fiber (glass) way more than for copper. And here you have a reason why toslink sounds like sh*t.
Summarized, I have never seen that isolation by glass sounded better. And many of us tried and spent a fortune (well, in the league of $1000). Adnaco stuff and such.

Peter

[jabbr]

Now if optical cables get impedance mismatches we have bigger problems

Hi Jonathan,

Yeah, you say it in a joking fashion, but I know someone in one of the states of your country who makes a living out of reflection testing of fiber cables (I have seen his gear).
I am not sure but I think fiber needs damping too (sunglasses ? haha).

Anyway, it is my kind of promise that only part of it is about the cable itself. The largest part is at the end point(s) where noise is re-generated. For fiber (glass) way more than for copper. And here you have a reason why toslink sounds like sh*t.
Summarized, I have never seen that isolation by glass sounded better. And many of us tried and spent a fortune (well, in the league of $1000). Adnaco stuff and such.

Peter

Well not all fiber is the same, and toslink is at least mostly plastic and the endpoints are nothing like those for Ethernet. Glass is clearly not "way more" than copper in terms of Ethernet noise. This has been extensively measured because, well, a little bit of noise and jitter can wreak havoc with a 10g or 40g or 100g Ethernet line... no problem over kms for singlemode Ethernet with really good lasers and transceivers. Eye pattern measurements show that.

I have no experience with Adnaco fiber USB units so can't say but yes they are expensive ... waaay more than fiber ethernet units that cost a fraction and have much better performance ...

Of course a high quality fiber cable is needed, but frankly let someone try and improve over Corning which is off the shelf.

[Telstar]

The interface has to be there, and the last time I looked into it (say a year ago or so) it did not exist. Oh, it exists, but by means of tricks I need to apply too. But two time the same trick on top of each other does not work.

Or put in another way : they must be natively 24/768.

Then the fpga needs to be programmed and a custom driver written. All current implementations support only 24/192 although multichannel. The bandiwth is way more than USB2 and is not a constraint.

I expected to see already something in this direction but nada, zero. The AES67 standard is 2 years old already. Yes, I'm talking about Dante/Ravenna here.
Alas, after 5 years (3 of which I have been away and the others moving) I expected to see usb3 receivers as well with multichannel DXD but I was wrong. There are none.

On paper AOIP is superior to USB, the cables do NOT resent as much from pc-induced noise, and it shouldnt be harder to implement.