EtherREGEN: The long development thread. [Some Gen2 dev. pics and update starting on page 92.]

[austinpop]

On 3/22/2018 at 12:51 PM, Superdad said:

the XO itself

 

Thanks for sharing!

 

Just curious - did John and you consider using XO or even OCXO? Was that a cost tradeoff?

 

Of course, I can wait until you're ready to say more about the clock design.

[austinpop]

4 hours ago, Superdad said:

 

Well the clocking in our switch is much more sophisticated than is typical as:

a) we have to pass the clock back across the isolation moat;

b) more than one frequency is used.

That is all I want to say about the clocking architecture at this time.

 

As for use of an OCXO: As John has written about this, cheap OCXOs are no better (often worse) than a good XO--one needs to spend hundreds (at OEM pricing) to get a really good OCXO.  And for data clocks (for USB, Ethernet, or processors), such is total overkill.  Save really great clocks for the DAC audio clock(s).  Even the clock boards in the SOtM units don't use a OCXO as a reference for the SiLabs synthesizer chip. 

Besides, this switch--and the issues we are tackling--is about far more than just clocking...

 

Hi Alex,

 

Well - I am going to defer to John's expertise on this. Major kudos to him and you for trying to demystify this space. The audiophile switch market is heating up, and it will be interesting to contrast what you guys come up with versus other approaches:

1. SOtM-modded switches, slaved to sCLK-EX on another box, in turn fed by an expensive OCXO 10 MHz reference clock
2. SOtM's impending standalone switch (details unknown)
3. Linear Solutions switch (OCXO in situ, plus external LPS)

I have no horse in this race. I just know how much I love the sound quality of config #1, which is what I currently have. If you guys can deliver that or better at a fraction of the cost, you'll have a major winner!

[austinpop]

4 minutes ago, Superdad said:

 

Well I might start a poll because it seems like half of you want 50 ohms and the other half 75 ohms.  We can do either--but not both.  It is simply a single tiny surface mount resistor change--and either the 50 or 75 ohm BNC.  But it is not cost effective to run both versions of the board--and rear panel labels (then get stuck with more of one versus the other), especially since 90%+ of buyers will never even be using it with an external clock.

 

Perhaps an impedance selector like this?

 

[austinpop]

28 minutes ago, Superdad said:

Given that all the major 10MHz reference clocks on the market have both 50 and 75 ohm outputs, why does it matter which we choose for input?

 

Come on... TedAlex. This is not an attack. We're just asking for a reasonable option. There are constraints for customers too.

1. older Cybersahfts have only 50Ω outputs
2. SOtM OCX-10 impedances have to be selected at time of purchase
3. Ref 10 has 6x 75Ω and 2x 50Ω

For any number of reasons, including the cables they own, customers may want to select the impedance of the input on the EtherRegen they buy. The competition provides that option too: http://www.sotm-audio.com/sotmwp/english/shop/snh-10g/

 

I know @JohnSwenson is more aware than laypeople that impedance matching involves BNC connectors as well as the internal and external cable, which is why we would trust you guys to do it right. Is an impedance switch such a bad idea? And it would save you from building and stocking 2 different versions.

 

If Cybershaft can implement an impedance switch, I'm sure John can too.

[austinpop]

5 minutes ago, Superdad said:

 

Well we were going to settle on 50 ohm, but it seems we can't please everyone. 

 

I can live with either impedance, just pointing out one size does not fit all.

 

Don’t shoot the messenger!

[austinpop]

50 minutes ago, Superdad said:

Some folks are today getting really hung up on this clock issue.  But what I have trying to explain is that our piece is about way more than clocking.  Perhaps I reopened the thread too soon because we are not yet ready to explain all of what makes what we are doing unique (and for propriety reasons we will have to decide just how much to reveal even when we do).

 

Very interesting Alex. I know some people get their knickers in a twist at the benefits from clocking stuff, and that's their problem.

 

I have never been wedded to a single approach, and have always gone where my ears told me. I have long felt that our empirical findings with better clocks (lower phase noise), better PSUs, and better isolation have all been related, and centered around eliminating analog noise that rides along the digital chain.

 

Therefore if you are hinting that John has had a breakthrough that ties these concepts together, and that he has engineered a solution that is immune to things like lower-phase noise external clocks, then color me intrigued! That would be revolutionary.

[austinpop]

10 hours ago, JohnSwenson said:

 

The problem is that in order to have both a 50 and 75 external input you need both a 50 and 75 BNC jack. There is no way to properly electrically implement both with one jack. Any such approach will degrade the signal, I have a feeling those that want to use multi thousand dollar external clocks will NOT want to have the circuit deliberately degrade the signal! Thus in order to do it we will need to have both jacks. The big problem is there is no room for both jacks on the case we have been planning for, to do it would mean going to a larger enclosure which will cost more money, not just for the extra parts, but for the enclosure and larger board.

 

There is a good possibility that traces on the board will wind up being longer because of the larger board, possibly slightly degrading performance.  Is it worth it?

 

My personal leanings are against it, when we go with a different case we have not used before, there are ALWAYS several iterations of case and board that have to be done in order to get everything just right. This adds a lot of development cost AND a couple months to the delivery time. Do you guys REALLY want that?

 

John S.

 

You're the manufacturer. It's up to you to decide what you can deliver, and what tradeoffs you need to make to do so.

 

Obviously, as customers, we want the best performance AND ultimate flexibility. That doesn't make us bad people. It's what customers do!

 

If you can deliver only one impedance, then pick one (take a poll, for example) and do it - and accept the fact that perhaps some customers will regard that as a showstopper. That's life. You're not going to able to please everybody.

[austinpop]

43 minutes ago, charlesphoto said:

Man, you clockers sure are a tough lot.  

 

The decision to offer a reference clock input was made unilaterally by Uptone. Nobody asked for, let alone demanded, it.

 

If they can deliver the best SQ among all the switches without it, I'd be more than happy.

[austinpop]

2 hours ago, Superdad said:

So here is how we are going to solve this edge-case issue:

a) When we have the first run of 250 boards made, I'll have our board house leave the BNC jack (and the one SMD resistor) off of a couple dozen boards.

b) The product will by default come with a 50 ohm BNC (I am not going to make every buyer choose because most buyers won't be using an external clock or even know what it is for).

c) In the fine print on the web page it will say to e-mail if you know you are going to need a 75 ohm clock input jack.  For those I will then insert a 75 ohm BNC and solder in the corresponding terminating resistor.

 

Perfect. Well done! 

 

I will probably hit you up for 75Ω when the time comes, because there are more choices of good 75Ω cables, and I have more 75Ω ports on my Ref-10. 

 

But choice is best. Life, liberty, and the pursuit of impedance. Wait... ?

[austinpop]

2 hours ago, Superdad said:

 

Thanks Jud.  The Banff Springs Hotel (first opened in 1888!) is where we are staying. It's very romantic, close to all three of the big ski parks, and Banff is a really cute town by the Bow river. My wife's birthday is also this weekend and since I am terrible at planning gifts, I'll likely just take her shopping.  I think she wants ski boots. 

 

 

I love this place. We stayed here 2 summers ago, and it was just gorgeous. It truly is a castle when you're inside!

 

Happy anniversary!

[austinpop]

44 minutes ago, Superdad said:

 

I often make typos in my posts and it is no big deal.  But I am sure glad I did not mess up that one! 

 

Yes, as a rule, your posts are not chock-a-block with typos.

[austinpop]

11 minutes ago, lmitche said:

Well yes, power quality is what I am after. The claim was that the EtherRegen was symmetric suggesting one side was as clean as the other. As I understand there is one power input, and there is galvanic isolation from one side to the other, I was curious if power quality is the same on both sides.

 

In other words, if I connect two devices, will it matter which side faces the DAC vs. the other?

 

Thinking further, and power aside, is the clocking symmetric?

 

I appreciate that this is mostly theoretical as no one has heard the ISO Regen yet. Some of this may be proprietary as well.

 

+1

 

I have been wondering the same. I certainly formed an impression during earlier discussions that the two sides of the isolation boundary were symmetric. Perhaps I misunderstood? I've not been following the back and forth in detail.

 

The scenario I'm thinking of is where the router attaches to the single port side, and then the other side has multiple clean ports for server, endpoint, etc?

[austinpop]

41 minutes ago, Superdad said:

 

That is correct.  I really need to update a number of things in the very first post of this thread. But hopefully it won't be too long before we start a fresh topic for the launch.  

 

 

Well on the 4-port side (5 if you could the SFP in the 'A'-side domain), a lot of thought has gone into the choice of magnetics and the switch chip.  In the expensive 4-port module we chose, each RJ45 port has 12-cores, whereas lesser ones will have only 4, 6, or 8 cores. And the switch chip we chose is an ideal match as its built in PHYs are set up for separate capacitor-coupling to ground of the center-taps of each port.  So there is quite good port-to-port isolation, thus there will be very little transference--of any leakage currents coming in on the cables of other gear--between these ports.

 

As for the clocking, we think the final domain out to the DAC-connected streamer/renderer is the one that matters, hence the single 100Mbps port on the actively isolated  'B'-side. Ours is the only dual-clock-domain switch, so it can't really be compared to how others do things.

As I mentioned before, some people are cascading two SOtM switches at great expense to achieve the reduction of leakage, and phase-noise overlay on the clock and ground planes that we will deliver in one much smaller package. 

 

 

Alex, I'm getting mixed messages here. Are the A and B side symmetric (in terms of SQ) or not? It sounds like you're saying that it's not, and for max performance/SQ,

• the A side should be used for the upstream (router, upstream switch, bridged server, etc)
• the single100Mbps port on the B side to the endpoint.

I recently posted a picture on another thread of my current setup, where I use 2 ports on what you would call the clean (or B) side: one to the endpoint/streamer, and one directly to the DAC. This enabled me to evaluate the benefit of the switch on an Ethernet DAC, and compare the Ethernet and USB inputs of the DAC, both paths benefiting from the switch. Here's the picture:

 

My thought on how to replicate this with the EtherRegen - in the fulness of time! - was to reverse the A and B side: connect the upstream to the single port on the B side, and then connect the streamer and DAC in my picture to 2 of the 4 ports on the A side.

 

It sounds like that is bass-ackward and would be a suboptimal use of the EtherRegen.

 

Please clarify?

[austinpop]

1 minute ago, lmitche said:

One could of course use 4 etheregens between 1) router and Nas 2) Nas and server 3) server and endpoint 4) endpoint and DAC!

 

No, thank you!

[austinpop]

40 minutes ago, JohnSwenson said:

Let me see if I can bring some clarity.

 

There are two types of "SQ degrading" influences the EtherRegen is designed to radically decrease, leakage, both high impedance and low impedance, and clock phase noise. The clock phase noise travels on the Ethernet signal itself (every edge coming out of any digital device caries the phase noise of the clock used to "clock out" that edge).

 

The very carefully chosen transformers on both sides play an important part in decreasing leakage. The active circuitry in the path across the moat adds a very major decrease as well. The result is that the leakage from A to B OR B to A is is decreased a huge amount. The decrease in leakage from one port to another on the A side is still quite significant but not nearly as much as when going from side to side.

 

The circuitry across the moat is designed to essentially eliminate the signal borne phase noise from one side to the other, it doesn't matter which direction, it works identically in both directions.

 

The circuitry between ports on the A side decreases these phase noise effects to some degree but not nearly as much as going from side to side.

 

There is ONE small difference between directions going across the moat: The clock generator is on the B side, so the circuits on the B side get a "pristine" clock. The clock from the B side goes through a very special isolator to the A side. This isolator has extremely low additive phase noise, much lower than any other isolator I could find. (it aint cheap!) The clock on the B side has slightly worse phase noise than the clock on the B side because of this. Whether this is going to be audible, who knows. Remember all the decrease in leakage and external phase noise is still there.

 

Going from port to port on the A side will be better than any other switch out there, but going from side to side (either way) will be a whole new world.

 

Because the B side has a slightly lower phase noise clock it is usually better to have the B side port connected to the streamer etc, but if you need to cross the moat the other way (such as using the SFP cage to drive optical into a streamer or DAC that has an optical input) That is also fine. The same decrease in leakage and external phase noise exists either way, the only difference is the slight increase in phase noise of the clock when going from B to A.

 

Because of this slight increase in phase noise when going from B to A, if you use a REALLY good external clock (such as a Ref10), you will only get the advantage of such a clock when connecting the B side to the streamer.

 

I hope this makes some sense.

 

John S.

 

Thanks John,

 

Now it all makes sense. Thank you for the cogent and detailed explanation. 

 

So it is indeed the case that theoretically, the B side is "cleaner," and the best use of this switch is to connect the upstream network to a single port on the A side, and the endpoint on the B side.

 

And if you're Larry, repeat this 4 times.

[austinpop]

3 minutes ago, Superdad said:

By the way, because the EtherREGEN is mostly symmetrical on both sides of its moat ... The 'A' side has four Gigabit copper ports and one Gigabit-ready SFP cage.  The 'B' side has a lone 100Mbps copper port.

 

So if you have multiple devices you want to connect on the clean side, you could run your "dirty" router to your B port.

 

Which presents a Shakesperean dilemma - to 'B' or not to 'B': that is the question. 

[austinpop]

20 minutes ago, Superdad said:

 

Aside from having just stated that there is not a "clean" side and "dirty" side, what you propose is quite the opposite of what we recommend.

The #1 consideration that every EtherREGEN user should start with is to attach the DAC-attached endpoint by itself to the 'B' side copper port.  

 

Second choice--only because multiple devices on the same side of the moat is not as ideal--is to feed the EtherREGEN from upstream network via its 'B' side port, and then attach your "audio" devices to the 4 copper ports (and 1 SFP port) on the 'A' side.

 

Hi Alex,

 

I hope you didn’t take what I wrote as a criticism? It wasn’t. Admittedly I enjoyed making a little joke, but it’s actually a serious consideration, and I wasn’t being gratuitously facetious.

 

Maybe because I’m a reviewer, I often have multiple DACs/endpoints on hand at any given point in time. For example, I am currently reviewing a piece through both the USB and Ethernet input. So I want to use the clean side of the “regen” switch to connect both to the the DAC-attached endpoint and to the DAC directly. And then there could be other endpoints.

 

So I can see a lot of value in using the multiport side of the switch as the clean end. 

[austinpop]

1 hour ago, Superdad said:

The main point I was trying to make to you is that putting an inferior switch between the EtherREGEN and whatever audio endpoints will degrade the signal integrity and pollute the clocking.

 

Now Alex, why on earth would I - of all people - even dream of doing that?

 

My point to you was that there are legitimate use cases that need multiple clean ports. Here's the picture from an impending review.

 

 

The EtherRegen could slot into where the SOtM switch currently is. Do note, from the picture: there is one "dirty" input, and 2 "clean" outputs. 

[austinpop]

1 hour ago, Superdad said:

 

LOL.   Of course I know you know that your's is something of an extreme case.  But considering you could almost purchase 3 EtherREGEN units for the price of one SOtM switch I am not too worried... 

 

Not that extreme. Just one EtherRegen could replace the SOtM switch in my picture, with the same PSU and the same 10MHz reference clock. A shootout just begging to be done.

 

Hint, hint.  

[austinpop]

Sorry to hear about the rib, @Superdad. Hope you feel better soon.

[austinpop]

3 hours ago, Nenon said:

 

I guess I need to explain how pre-loading in RAM is different than streaming.

 

Not all the DSD files are that big, but some are. Let's take an extreme case. It takes over 80 seconds to transfer a 1GB file over 100 Mbps connection. So let's say hypothetically that I have an album with 10 DSD tracks of that size. Here is what happens:

- I click play.

- I wait 80+ seconds for the file to load in RAM. 

- The track start playing. The music stops at the end of the track.

- The second track starts loading.

- I wait 80+ seconds for the second track to load.

- Repeat the same process above for the rest of the tracks... 

 

1 GB file is a very extreme case. Most of my files are much smaller. Let's say an average file is 150 MB. That is still a 12+ seconds gap between tracks. That would be quite an unpleasant experience over 100 Mbps connection. If we bump it up to 1 Gbps connection the gap would be just a little over a second on an average file size. 

 

Interesting scenario. I certainly have tracks that exceed 1GB in size. Are you referring to Stylus on Euphony OS - i.e. "Buffer before playing = 100%"?

 

If so, while it is definitely true you'll incur the pre-load delay for the first track, it should not be an issue for subsequent tracks. Especially if you've also set the "Use Cache"  flag. In that case, all tracks that are added to the Queue start getting loaded into the local filesystem cache, from where they are then loaded to RAM for playback. While you'll wait for the first track to transfer, the remaining tracks will transfer while you're listening. Additionally, the gapless playback algorithm in Stylus preloads the next track (from cache), so you really don't have to incur the delays on every track.

 

That said, another place I've found 100Mbps to be limiting is during system upgrades, especially if it involves a full OS upgrade of the DAC-attached endpoint/server.

 

Presumably, if you keep the NAS and the music server on the same side of the moat, while you don't get isolation, I'm guessing you still get reclocking?

 

Good question.

[austinpop]

1 hour ago, Nenon said:

Thank you, John. Will patiently wait and try when the time comes.

 

+1. That's exactly what I was thinking too. I have witnessed similar sound quality degradation simply from scanning my local library, even with the network cable unplugged.  So I agree that it's more likely the result of the interrupt processing. 

 

Let's move on. We'll post the outcome of our tests in another thread when we have an EtherREGEN in our hands. 

 

Indeed. 

 

Nenon, if you're interested, I looked into this interrupt behavior in the context of another workload: Squeezelite with large buffers on AL in ramroot. Different code, but similar behavior on the network. Here's the link: https://audiophilestyle.com/forums/topic/54933-audiolinux-and-nuc-troubleshooting-and-tuning/?do=findComment&comment=904234

 

 

Let's followup on that or the Euphony thread, as it's definitely OT here.

[austinpop]

11 hours ago, Aidagent said:

This is my first post in this forum. First I want to tell you that English is not my native language. I hope you have patience.

 

I have followed this thread for a long time with interest. I admire enthusiasts who put their soul into creating the best conditions possible to reproduce the music as it was created. I am impressed by what you create and wait patiently for the product to come into my setup. I think it's good that you take the time you need to make you feel satisfied and ready to launch.

I have several products in my setup that are made by enthusiasts whose goal is greater than just making money. It is nothing wrong in makeing money. but if it is the highest goal, it often leads to short-term decisions that rarely yield the best results in the long run. I am not technically savvy. so it is difficult for me to judge how great the product will do in my system. But every litle bit of disturbance thats cleaned out of the signal adds up to a better musical experience.

 

Thank you for your work and the way you communicate with us followers. And thanks to everyone who writes in the thread for your interesting questions.

 

Johnny Cash Johnsson (without cash)

from Sweden.

 

Welcome to the forum. I like your signature, Johnny Cash (without cash)!

 

I’m in agreement with @Superdad - your English is probably better than half the English speakers who post here.  

[austinpop]

1 hour ago, Superdad said:

Hi Steve. Thanks for asking.  Most all the pain is gone.  I still feel it a bit when I sneeze hard, and I can not yet take extremely deep breaths.  Also, I have mostly stopped sleeping on my left side, not because of much rib pain, but because I wake up with a lot of pain in my shoulder and under my shoulder blade if I do.  I think this must be somehow related to the rib fractures. There is an experienced bodywork therapist I sometimes use--and he is about to get a call from me.

 

I hear ya. Rib fractures aren't what they're cracked up to be. And they're certainly nothing to sneeze at.   I hope I didn't make you chuckle too hard.

 

Feel better soon. And great to hear about the EtherRegen progress.

[austinpop]

5 hours ago, Superdad said:

I am really not happy with the tone of recent posts here guys.  

 

A bit silly, ya think?