UpTone Audio EtherREGEN (Objective Discussion Only)

[jabbr]

7 hours ago, Superdad said:

 

Except that the funny thing is, while the Audio Precision Amir worships at is a very nice general purpose unit, it is not that well suited to measuring jitter in DACs. What’s that? Heresy!  Yet if you look at the specifications for the ADC in the top-of-the-line APx555, you will see that its own jitter is 600 picoseconds!  So that will swamp the jitter details of a device under test—which might be in the range of 10s of picoseconds--or ideally less.

 

Does anyone measure jitter on an AP, I mean really ... you are using a 1Ghz+ scope to measure jitter on a 25-100 Mhz DAC clock right?

 

7 hours ago, Superdad said:

 

As I've said elsewhere, John has already measured these effects--with a Wavecrest system better suited to this--at the clock pin of DACs. 6dB improvement at 10Hz offset is what he has seen so far with EtherREGEN. Our tests using a custom 32 bit high-speed SAR ADC and analysis s/w--to show things after the DAC will come later. 

 

Right, I mean in all fairness, the AP is designed to measure analog signals, not high speed digital...

 

and so ... I can pretty much guarantee that you aren't seeing a 6 dB improvement compared to my 10Gbe switch, and certainly not compared to my 100Gbe switch in which the jitter budget is zero ... also both of these systems have been tested NOT TO PROPAGATE JITTER ... its just not allowed in modern Ethernet. I mean fair enough that some $12 switch is built like crap.

 

Here are some jitter budgets:

https://www.sitime.com/company/news/blog/what-you-need-know-about-phase-noise-and-jitter-high-speed-systems

note that 100Gbe (4x25) is ) 0.061 picoseconds  

 

In all fairness to the EtherREGEN, my professional switches retail for >$30k (I paid a very small fraction). But they have dual SMPS PSUs and similarly my 100Gbe NICs are also powered off a bod standard server PSU via the PCIe bus...  You can't even measure the low jitter (<  61 femtoseconds) on these puppies  

 

Ok, so let's forget this TP-Link or Trendnet whatever generic ancient spec Ethernet stuff, and at the very least compare against a very old Cisco device (I mean assuming you are looking at 100m Ethernet which spec is circa 1995 and is officially on "Legacy" status with the IEEE)

[jabbr]

5 hours ago, opus101 said:

I read that - what's missing for me is the mechanism by which the groundplane noise gets to affect the clock. If the designers put the clock itself on the same groundplane as the rest of the logic I can see how it would matter but designers who're aware of this issue would likely use a separate groundplane (island) for such sensitive circuits as an oscillator.

JS is certainly correct that ground plane noise affects crystal oscillators. This is a very interesting issue. 
 

There is a nonlinearity whereby noise at f causes f frequency offset phase noise. 
 

ie noise at 1Hz causes phase error f +|- 1 Hz and 10 Hz causes phase error f +|- 10 Hz ... this is described in Rubiola’s textbook: https://www.cambridge.org/core/books/phase-noise-and-frequency-stability-in-oscillators/445C12C4ECBFCD7765116E61561EC0FE

[jabbr]

25 minutes ago, Jud said:

 

I would be interested in exploring whether there is any objectively measurable reason behind my listening preference for the ER vs. Cisco 2960, and for the ER by itself vs. the 2960 followed by the ER. While the "bog standard" SMPSs may not cause jitter or noise over Ethernet, I wonder what they may do wrt noise running over system component ground planes.

 

I don’t have a Cisco 2960 but presumably JS could test this against the EtherREGEN and show us if the DAC clock has less jitter. 
 

There are SMPS and there are SMPS ... I’m saying that the Mellanox SMPS obviously causes <60 femtosecond jitter in the 100Gbe switch ... and they are at 400Gbe so imagine how low jitter those systems are and yet still powered by SMPS  

[jabbr]

3 hours ago, Jud said:

While the "bog standard" SMPSs may not cause jitter or noise over Ethernet, I wonder what they may do wrt noise running over system component ground planes.

You can also measure the spectrum of noise on the ground plane — this is easier but use a spectrum analyzer that can do sub-Hz measurements. eg 0- 10kHz or so ....

 

Thats a simple measurement with the right analyzer.

[jabbr]

7 hours ago, cat6man said:

 

wish i had a 100Gbe switch 

 

learning Sonic — Linux for switches  

Quote

 

is there a similar jitter spec that 1Gb and 100Mb switches must meet?

 

Not really, the 10GBase-X specs were the first to define end to end conformance testing (using eye patterns) which means everyone needs to play nice together — obviously if jitter were additive, then serial hops would blow the eye pattern — thankfully the Internet works  

 

Again there is a role for EtherREGEN as a high quality slow Ethernet switch. If anything avoiding common mode noise may be more important that jitter — lets see!

 

Quote

does anyone know what typical Fios router/switches have for jitter?

 

and to follow up on superdad's comment above, does the ASR's a/d converter really have a 600ns jitter spec and, if so, does that invalidate the ASR measurements of jitter? 

There are different ways to measure jitter — eg  looking for pure tone widening , but again I’d simply look at the ground plane noise spectrum because if the switch doesn’t dump noise on the receiver ground plane hen the argument is moot, or if it does then whether by reducing common mode noise or jitter is subject to further measurements. 

[jabbr]

@JohnSwenson: I read your white paper.
 

You have identified the well known nonlinearity as described by Rubiola whereby power supply noise at eg 10Hz causes a frequency offset error in a crystal at f +|- 10Hz. 

 

Ok that’s all correct, however:

 

There is not a well identified inverse nonlinearity whereby phase error an an input f +|- 10 Hz is downconverted to ground plane noise at 10Hz. 
 

If there is switching noise from a gigahertz input Ethernet signal, it will cluster at a gigahertz. 
 

On the other hand common mode noise transmission down a cable could cause 10Hz (or 60Hz) noise in a receiver. 
 

There is no known mechanism to say that 10Hz noise transmission by a gigahertz Ethernet cable has anything to do with Ethernet clock jitter as opposed to common mode noise..

[jabbr]

7 hours ago, vortecjr said:

If you want to learn more have a look at this short informative video on this subject.

 

 

 

We mixing up what is being tested: there is no one type of jitter and no single way of testing it. The above discussion regards jitter on an SPDIF interface and no question that topic is of interest to DACs.

 

Regarding the EtherREGEN, the jitter being discussed, and which @JohnSwenson discusses in the whitepaper, is jitter on the Ethernet interface: These are drastically different things!

 

Let's start with a few questions:

1) Does common mode or differential mode noise  on an Ethernet interface affect the downstream DAC, i.e. does it pass through one of the Rendu devices? 

2) Does jitter/phase noise in the Ethernet clock itself affect the DAC?

 

1) There is a clearcut electrical mechanism whereby common mode noise causes ground plane noise (e.g. leakage currents) and this affects audio circuits -- eg ground loops. I will buy a measurement of reduced ground plane noise with the EtherREGEN compared to xxx switch as indeed reduction in common mode noise.

2) There is not a clearcut mechanism whereby Ethernet clock jitter causes significant ground plane noise in the audio spectrum. I don't buy, without substantial additional testing, that a reduction in ground plane noise in a DAC results solely from reduction in Ethernet clock jitter.

 

The AP device is not, to my knowledge, capable of measuring Ethernet clock jitter -- can it create a 1 Gbe eye pattern?

 

Have you measured the jitter/eye pattern on the opticalModule? (obviously this will block common mode noise  ) I'm going to assume the AP is not capable of measuring this.

 

 

[jabbr]

3 hours ago, Superdad said:

 

John's Lecroy WavePro 7300 3-GHz scope can create and show Ethernet eye patterns, though just the nature of an eye-pattern (build up of thousands of sweeps) makes me wonder if it would reveal anything useful.  But I'm not the expert on measurement...

Eye patterns don’t tell everything, that said your shiny new phase noise analyzer doesn't measure phase error at 1Ghz. The phase error plot is statistical regardless. 
 

The newer Ethernet standards, which are specifically designed so that jitter is not additive across hops, specify eye pattern compliance testing. They specify a “stressed eye pattern” test whereby a maximally poor sender is tested against a receiver. If the receiver passes, then regardless of how bad the sender is (within compliance) the receiver’s jitter is not additive, nor is the next hop. 

[jabbr]

2 hours ago, vortecjr said:

Come on...you know the answers to these questions. In regards to the Rendu John advised me that he would use one in the measurement because of it's design. I'll ask John to explain why he is using a Rendu so you guys know his reasoning. My AP does not measure eye patterns. AP don't seem all that interested in USB and Ethernet.  

Of course, and I’ve said that the AP is not appropriate to test the Ether REGEN’s jitter — it’s not designed to do so, regardless of the fact it can analyze SPDIF .

 

Now the j-test is a reasonable way to test the output of a DAC and the AP is reported to be excellent audio analyzer. 
 

That said the specific claims about how jitter on the Ethernet network affect the DAC make specific predictions regarding appropriate measurements. Let’s see them. 

[jabbr]

3 hours ago, vortecjr said:

I spoke to John a while back and he mentioned someone had done some research into all this. Not sure if it was Rubiola or not.  

Rubiola has published papers (with co-authors) and written a textbook (which I have). The effect of power supply noise on crystal oscillator phase noise is well described. 1/f noise causes 1/f-offset phase noise. The inverse has not been described. 
 

I am willing to look at relevant measurements and if @JohnSwenson can demonstrate 1/f-offset phase noise resulting in 1/f noise on the receiver ground plane then I will gladly accept his argument. That’s an eminently testable statement — I’d use fiberoptic Ethernet to eliminate the confounding common mode noise variable. 

[jabbr]

4 minutes ago, opus101 said:

What is it that makes the groundplane noise at the receiver relevant? Can't we just check at the output of the DAC (I'm assuming here we do have sensitive enough measurements for low-frequency noise)?

The proposed mechanism specifically predicts noise on the ground plane.

 

Ive given a very specific measurement result which would convince me, do you have a different one? What and why?

[jabbr]

30 minutes ago, opus101 said:

So what is it that makes this mechanism - even if shown to be active in practice - relevant in the bigger picture?

the 1/f -> 1/f offset is known, if the inverse were the case (no evidence that it is) then there was s a mechanism whereby a “jitter signature” could pass through a switch. 
 

My hardware is certified that this is irrelevant — either nonexistent or insignificant, but 1G or 100m hardware is not required to meet end to end jitter standards. 
 

The entire argument that server noise makes it across the Network depends either on this or common mode noise transmission. 

[jabbr]

18 minutes ago, Audiophile Neuroscience said:

 

 

As I see it, whether at the receiver or output of the DAC, ultimately one must then demonstrate its correlation/concordance with audibility. First things first and that is to demonstrate an objective measurable phenomenon related to signal integrity that might be relevant. Yes, you can argue that a change in one part of the signal path may be more convincing than another but discoveries are often a matter of piecing together many parts of a puzzle.

Marketing is about telling a story, and this story has resonated at AS for years. Whether the proposed effect is audible depends on whether it actually exists in the physical world. If it doesn’t exist then we can look for another story. If the effect exists but it’s not audible then it might be audible in other circumstances. My bias is that this is all about common mode noise rather than jitter, so let’s see if my story is more relevant (I’m not selling anything though 😂)

[jabbr]

1 hour ago, Superdad said:

 

On 3/13/2020 at 10:07 PM, jabbr said:

The proposed mechanism specifically predicts noise on the ground plane.

Yes, and we can and have measured that.  And when John's expensive new Jackson Labs PhaseStation system is delivered and set up he will be able to present DAC-clock-pin jitter comparisons to demonstrate the phase-noise effects.

 

 

Since you've measured that, and haven't provided measurements, I have to assume that your measurements do not support your proposal that Ethernet phase error f +|- x causes receiver ground plane noise at x frequency.

 

i.e. assuming your ethernet clock is 25 Mhz, your proposal suggests that phase error 25 Mhz +/- 10 Hz results in ground plane noise at 10 Hz ... right? Can you demonstrate that? It is a very specific prediction.

[jabbr]

1 hour ago, manueljenkin said:

Got carried away. Was talking about usb packets. Not ethernet.

 

That's ok, hardly anyone is talking about Ethernet here 😂

[jabbr]

4 hours ago, The Computer Audiophile said:

Ethernet is not jitter-free. Fact. You can't weasel your way around that.

 

Yes we should qualify our statements because nothing (sic) in the real world is ever jitter free or noise free or perfect or true or false.

 

What we can say is that jitter is not additive across modern ethernet interfaces and jitter signatures do not pass across modern ethernet interfaces. That's true for modern ethernet interfaces which are tested compliant with modern ethernet standards. Let me be very specific here, if the Ethernet interface jitter in a stressed eye pattern test were additive, the interface would fail the test. Asked and answered.

 

The EtherREGEN is not compliant with these modern standards (I assume the testing has not been done, and these standards are 10GBase-X, 40GBase-X, 100Gbase-X 200, 400...), yet legacy ethernet is in common usage among home audio equipment. It indeed serves a niche for home audio networks and equipment using legacy protocols.

[jabbr]

1 hour ago, austinpop said:

Since this is the objective forum, this thread should just go on an extended hiatus until John publishes his measurements. What else - objectively - is there to discuss?

Objective discussions of papers include objective discussions of the statements made in the papers. The paper has no measurements rather proposes a theory to explain a theoretical electrical phenomenon. Publishing a paper invites discussion of the theory. 
 

According to your definition I am a subjectivist yet I enjoy discussing the scientific merits and engineering. I’m quite capable of listening for myself and judging what I like, that said life is short.

[jabbr]

@Jud its great that having a special switch is making you happy! The best thing about all this it that the nonsense about a direct PC to endpoint/DAC connection and using the PC as a router is something we are hearing less about.

 

For folks who aren't inclined to get involved with networking hardware (not saying you), its great to have a pre-packaged "audiophile" solution. If this switch is complementary to the microRendu, then its seen as an all in one solution. In the same way the branding of the "systemOptique" with cables and preselected SFDP switches makes it easy for folks to plug and play.

 

(this is a non objectivist offtopic post, sorry)

[jabbr]

18 hours ago, alfe said:

My apologies Alex for my poor joke. 

That was actually very funny. I don't think @Superdad should take offense to that, I mean in all seriousness, the thesis is that jitter infects audio systems and is transmitted from component to component. 

[jabbr]

43 minutes ago, sandyk said:

 

 

 DBTs are frequently demanded by Objectivists. Are a tool used by Objectivists, and rarely by Subjectivists other than to satisfy the demands by Objectivists.

im not sure if I am objectivist but I haven’t selected my network equipment on the basis of DBT.

 

I have absolutely no uncertainty that the EtherREGEN is the best switch designed for audiophiles in the 100m speed range and <$1k new.

[jabbr]

3 hours ago, Seraph said:

So what is it, do you know what makes the device technically superior to any other 20 times more affordable switches

 

I think it should be obvious that switch cost is not an important factor here!

 

No one has shown evidence that the EtherREGEN isn't a perfectly capable low port count 100m/1g level 2 switch.

 

There are, objectively, many audiophiles who are looking for devices designed for audiophiles whether or not such devices have measurements. We have evidence of many audiophile products which are successful despite measurements. I think that its probably that some consumer devices including "cheap" generic Ethernet switches, might emit common mode noise, this has certainly been measured for the very cheapest wall wart SMPS power supplies. So if you factor a small run design tailored at reducing noise as well as an audiophile multiplier, the cost is reasonable.

 

You know it might be like the sofa I'm sitting on right now. Its not measurably better than many other sofas nor is it the cheapest sofa. 

[jabbr]

35 minutes ago, pkane2001 said:

Why?

 

Well I agree in this case, for example, the $$ I spent on the Topping D7s based on measurements, I found to be a big waste because I don't like the sound. On the other hand the Pro-ject S2D was a great purchase. Do you have a measurement that predicts my preference?

[jabbr]

6 hours ago, Audiophile Neuroscience said:

In essence this is what has been done.People have listened and made observations. Measurements are apparently pending.As pointed out by @Iving the process is a very complicated one from an objective scientific viewpoint.

 

There are some very simple measurements that could be made ...

 

For example:

1) noise on the ground plane of the EtherREGEN vs Cisco vs Trendnet

2) noise on the receiver ground plane e.g. microRendu using EtherREGEN vs Cisco vs Trendnet

 

None of that needs very fancy equipment to measure if one wanted to do the measurement.

[jabbr]

7 minutes ago, Superdad said:

 

Very much along the lines of what we will publish. Though as you know, it takes careful attention to environmental factors to properly measure very low level noise.  And the perturbations we are looking at do not need to be very large to have the effect on clock threshold jitter that we believe are the root cause of the sonic differences heard.

 

I am assuming that there should be a rather large and easily identifiable difference in the ground plane noise between your device and a generic low cost switch. 

7 minutes ago, Superdad said:

 

@JohnSwenson's expensive new PhaseStation is showing just how much environmental factors influence low level phase-noise measurement. DC cables, vibration, fields in the air, lighting, even body presence are causing wild wiggles at the levels he is testing. He is building cases and supplies to reduce those distractions. And we are not even talking about your favorite, 1/f noise. 

I hope that some easy measurements would convincingly show something, rather than chasing ghosts. 

[jabbr]

59 minutes ago, Audiophile Neuroscience said:

I would welcome these or any other relevant measurements. The keyword here I guess is relevant. For me, I would be interested in any measurement of the audio signal that shows a difference.  I do agree however with @Iving that the process does not stop there. Indeed I would anticipate that others will argue that the process does not even start there and specifically, the measured change in the audio signal must be at the output of the DAC, anywhere else being irrelevant.

 

No doubt that measurements at the DAC output are what is ultimately desired however there are things about network/usb phase noise that have been bandied about for about 5 years now without even the most basic measurements.

 

My experience is that when someone is having trouble demonstrating something like this with measurements, it means that whatever measurements have been done don't support the theory.

 

My own experience with audible ground loops is that they are easy to measure. We aren't looking for Bs meson's here  

 

I am eager to be ediucated.