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

[jabbr]

4 hours ago, R1200CL said:

My understanding is there isn’t any difference in using optical vs RJ45 A-side and then cross over to B-side. 

Not sure exactly what you are saying here and I’ve answered in the thread I posted on. I don’t know the nitty gritty circuit details — haven’t seen a schematic, but from what I understand, the EtherRegen is designed primarily for optical or copper in and copper out to the endpoint connected to the DAC. The fiberoptic link is already completely common mode isolated. 

[Superdad]

3 hours ago, nonesup said:

Hi Alex.

 Some answer for my previous post.

 

Are you thinking about a new USB Regen based on what you have learned with the development of EtherRegen?

😁

 

We are always thinking and working on all sorts of new products.  John and I have at least 3 years of new products mapped out and in various stages of design (ranging from proto-boards to just circuit sketches we know would be great).  But we share our plans cautiously and strategically. I'm sure you understand.

Thanks for your interest though...  

[jabbr]

3 hours ago, Superdad said:

 

We use a Crystek CCHD-575 at 25MHz as the reference to drive a particular Silicon Labs jitter-attenuating differential clock synthesizer (which we also have set to alternatively accept a 10MHz external reference clock).

We use all 4 available outputs of the synth, programmed to generate two separate 25MHz lines (for the Ethernet chips on both sides of the moat), and two 250MHz lines—for the high-speed, ultra-low-jitter differential reclocking flip-flops on both sides.  

 

Very nice. Those silabs chips are very slick.

Um, seriously, why do you enable external clocks aside for marketing to the many folks who don’t understand clocks or what you guys are doing?

A slight point — the output of the silabs doesn’t equal an independent clock —those clock domains on each side are synchronous — which is good

[LowMidHigh]

33 minutes ago, jabbr said:

 

Very nice. Those silabs chips are very slick.

Um, seriously, why do you enable external clocks aside for marketing to the many folks who don’t understand clocks or what you guys are doing?

A slight point — the output of the silabs doesn’t equal an independent clock —those clock domains on each side are synchronous — which is good

 

1) Nice of you to presume that many folks don't understand clocks.  

2) It's been explained many a time that the BNC connection is built-in to the chip, so allowing external clock is trivial both in terms of cost and labor. Further, the competing product in the marketplace offers that same feature, so why not compete head-to-head, to the benefit of people like you, who do understand the value of an external clock? 

[Superdad]

12 minutes ago, jabbr said:

Very nice. Those silabs chips are very slick.

Thanks.  And yes, they are slick--especially if one chooses the expensive ultra-low-phase-jitter versions. 

 

12 minutes ago, jabbr said:

Um, seriously, why do you enable external clocks aside for marketing to the many folks who don’t understand clocks or what you guys are doing?

You are right, inclusion of the external clock input was mostly just a marketing checklist item.  A page ago I explained it thusly:

 

An external reference clock is NOT required to get stellar performance from the EtherREGEN! 

It already has built-in the lowest-phase-noise production oscillator on the market (at least one where the OEM cost is not hundreds of dollars!)--positioned within millimeters of the chips that use it.  Plus virtually all the clock lines on the board are run as differential pairs--including across the moat--and the ultra-low jitter differential flip-flops we use are unparalleled in the world.  So real-world jitter is going to be far lower than any other switch with just a fancy clock board wired in. 

 

So why is there a BNC jack and support for an external 10MHz reference clock? Because it was easy to include (for probably a total of less than $10 more) as a competitive "checklist" item.  It is no secret that we are going to compete with the nearly 3x the price and 4x the size SOtM sNH-10G.  I'll skip all the in-our-favor technical advances comparison for now.  But since the SOtM sNH-10G has (if you pay the extra $200 and get the $1,700 version) a BNC for external clock; and since some of the enthusiasts here happen to have a fine expensive reference clock such as the Mutec REF10--it seemed like a good idea to include provision for that on the EtherREGEN.

 

Will use of an external clock with the EtherREGEN make any worthwhile sonic difference?  I have no earthly idea!  I guess some of you will find out.

 

 

12 minutes ago, jabbr said:

A slight point — the output of the silabs doesn’t equal an independent clock —those clock domains on each side are synchronous — which is good.

 

Yep, and while we were sort of expecting to have to program in a slight phase delay between the clock outputs for the flops, it turns out that the 10GHz-capable flops we are using have such a tight timing window between hold time and setup time that the chances of landing in the spot is infinitesimally small. (I am not explaining this clearly but I guess you know what I mean.)

[justubes]

Thanks John,

 

Also is the masterclock input 50 or 75 ohm per Silabs specifications?

 

Does the clock input accept any other frequency besides 10mhz. For example 5mhz, 25mhz?

 

Having considered the technicals and purpose of the isolator between  A and B ports,

Is the 10m also isolated by the moat if the clock input is also linked to the dac, this might cause a common ground contamination.

 

 

[Superdad]

2 hours ago, justubes said:

Thanks John,

I’m Alex, not John.

 

Quote

Also is the masterclock input 50 or 75 ohm per Silabs specifications?

We are choosing to use a 75ohm BNC and approriate termination resistors (50ohm will be available as special order).  Specifications of the clock synthesizer have nothing to do with the input impedance of the external clock connection.

 

Quote

Does the clock input accept any other frequency besides 10mhz. For example 5mhz, 25mhz?

No, we are programming it for 10MHz as that is the frequency of virtually every available external reference clock. To set it for some other frequency would not make any sense.

 

Quote

Having considered the technicals and purpose of the isolator between  A and B ports,

Is the 10m also isolated by the moat if the clock input is also linked to the dac, this might cause a common ground contamination.

Interesting question. John might have some other thoughts on this but I can state what I know:

 

We have positioned the clock synth (and the Crystek 575 which for 95%+ of EtherREGEN users will be the internal reference clock driving it) on the ‘B’ side (the one with the lone port). The external BNC port also comes in on the ‘B’ side of the moat. 

 

So while the clocking to ‘A’ and ‘B’ sides is equally ultra-low-jitter on both sides (remember all clocking runs on very tightly controlled differential lines), the power/ground domain the external clock connects to is the ‘B’ side (refer to the board photo I posted).

Thus if using an external reference clock it is probably better to also be using—as most people will typically—the ‘B’ side port for connection to the furthest downstream, DAC-attached renderer/endpoint.  Otherwise there is a chance the galvanic isolation will be defeated. It depends.

 

Thinking about this makes me wonder about external reference clocks which typically have multiple ports (I sure can’t imagine someone buying a $3,500 ref clock just for an EtherREGEN!).  Are the output grounds of those ports common to each other? This is an area I don’t yet know much about.

 

I suppose if common ground domains (and possible power supply loops between a clock and upstream gear) are an issue, it is possible that performance may be better just using the EtherREGEN’s internal reference clock. 

[That may be the case anyway since jitter/phase-noise at the end of a long clock cable may be worse than from the internal clock.]

 

Please be careful about getting too hung up on ultimate clocking—whch in fact our circuitry is already very close to.  We think the “magic” of the EtherREGEN’s performance will come from the total leakage blocking of the active-isolated, dual data/power/clock domains—plus all the care we put into choosing magnetics, voltage regulators, logic chips and all the differential signaling.

As I’ve said, this is an Ethernet switch unlike ANY other. 

[justubes]

1 hour ago, Superdad said:

I’m Alex, not John.

 

We are choosing to use a 75ohm BNC and approriate termination resistors (50ohm will be available as special order).  Specifications of the clock synthesizer have nothing to do with the input impedance of the external clock connection.

 

No, we are programming it for 10MHz as that is the frequency of virtually every available external reference clock. To set it for some other frequency would not make any sense.

 

Interesting question. John might have some other thoughts on this but I can state what I know:

 

We have positioned the clock synth (and the Crystek 575 which for 95%+ of EtherREGEN users will be the internal reference clock driving it) on the ‘B’ side (the one with the lone port). The external BNC port also comes in on the ‘B’ side of the moat. 

 

So while the clocking to ‘A’ and ‘B’ sides is equally ultra-low-jitter on both sides (remember all clocking runs on very tightly controlled differential lines), the power/ground domain the external clock connects to is the ‘B’ side (refer to the board photo I posted).

Thus if using an external reference clock it is probably better to also be using—as most people will typically—the ‘B’ side port for connection to the furthest downstream, DAC-attached renderer/endpoint.  Otherwise there is a chance the galvanic isolation will be defeated. It depends.

 

Thinking about this makes me wonder about external reference clocks which typically have multiple ports (I sure can’t imagine someone buying a $3,500 ref clock just for an EtherREGEN!).  Are the output grounds of those ports common to each other? This is an area I don’t yet know much about.

 

I suppose if common ground domains (and possible power supply loops between a clock and upstream gear) are an issue, it is possible that performance may be better just using the EtherREGEN’s internal reference clock. 

[That may be the case anyway since jitter/phase-noise at the end of a long clock cable may be worse than from the internal clock.]

 

Please be careful about getting too hung up on ultimate clocking—whch in fact our circuitry is already very close to.  We think the “magic” of the EtherREGEN’s performance will come from the total leakage blocking of the active-isolated, dual data/power/clock domains—plus all the care we put into choosing magnetics, voltage regulators, logic chips and all the differential signaling.

As I’ve said, this is an Ethernet switch unlike ANY other. 

Hi Alex,

 

Sorry for the mixup.

 

Yes, i will check, but am quite sure the Mutecs ar3 common. It will possibly defeat the fibre isolation as it will still be grounded to the sitry network side. Ismthe grounds of the B side totally isolated from the A and power ground. Somehow, with a Mutec clock this may not work out so well form the total isololation point of view.

 

There are clock like the new Cybershaft and Esoteric master clock which can have individual outputs ground lifted.

 

The cchd 575 have excellent phase noise, but theres also the non techinal specs of clocks that no one understands affects the sonis largely as well.

 

The reason is if the input master clock can be specified when  ordering? This will allow also the use of say the Mutec mc3+ reclocker for those who have that piece to utilise together with the etherregen.

 

 

[lmitche]

7 hours ago, Superdad said:

I’m Alex, not John.

 

We are choosing to use a 75ohm BNC and approriate termination resistors (50ohm will be available as special order).  Specifications of the clock synthesizer have nothing to do with the input impedance of the external clock connection.

 

No, we are programming it for 10MHz as that is the frequency of virtually every available external reference clock. To set it for some other frequency would not make any sense.

 

Interesting question. John might have some other thoughts on this but I can state what I know:

 

We have positioned the clock synth (and the Crystek 575 which for 95%+ of EtherREGEN users will be the internal reference clock driving it) on the ‘B’ side (the one with the lone port). The external BNC port also comes in on the ‘B’ side of the moat. 

 

So while the clocking to ‘A’ and ‘B’ sides is equally ultra-low-jitter on both sides (remember all clocking runs on very tightly controlled differential lines), the power/ground domain the external clock connects to is the ‘B’ side (refer to the board photo I posted).

Thus if using an external reference clock it is probably better to also be using—as most people will typically—the ‘B’ side port for connection to the furthest downstream, DAC-attached renderer/endpoint.  Otherwise there is a chance the galvanic isolation will be defeated. It depends.

 

Thinking about this makes me wonder about external reference clocks which typically have multiple ports (I sure can’t imagine someone buying a $3,500 ref clock just for an EtherREGEN!).  Are the output grounds of those ports common to each other? This is an area I don’t yet know much about.

 

I suppose if common ground domains (and possible power supply loops between a clock and upstream gear) are an issue, it is possible that performance may be better just using the EtherREGEN’s internal reference clock. 

[That may be the case anyway since jitter/phase-noise at the end of a long clock cable may be worse than from the internal clock.]

 

Please be careful about getting too hung up on ultimate clocking—whch in fact our circuitry is already very close to.  We think the “magic” of the EtherREGEN’s performance will come from the total leakage blocking of the active-isolated, dual data/power/clock domains—plus all the care we put into choosing magnetics, voltage regulators, logic chips and all the differential signaling.

As I’ve said, this is an Ethernet switch unlike ANY other. 

While I know this isn't the place, I have never seen a discussion of the impact of the hub based clock cabling schemes on Galvanic Isolation. It would be interesting to learn the facts about this once and for all.

 

Maybe this is a good "Novel" thread topic as many there have external clocking experience.

 

 

[marce]

16 hours ago, LowMidHigh said:

 

1) Nice of you to presume that many folks don't understand clocks.  

2) It's been explained many a time that the BNC connection is built-in to the chip, so allowing external clock is trivial both in terms of cost and labor. Further, the competing product in the marketplace offers that same feature, so why not compete head-to-head, to the benefit of people like you, who do understand the value of an external clock? 

External clocks are not the best idea for domestic audio, the extra length of cable causes more issues than it solves.

[Superdad]

8 hours ago, justubes said:

Yes, i will check, but am quite sure the Mutecs ar3 common. It will possibly defeat the fibre isolation as it will still be grounded to the sitry network side. Ismthe grounds of the B side totally isolated from the A and power ground. Somehow, with a Mutec clock this may not work out so well form the total isololation point of view.

 

I think you are overthinking this.  Please refer to what I has posted previously.  Ours is the only switch to offer isolated dual-data/power/clock domains anyway, so it is not as if potential defeating of the isolation by introduction of "grounds" common to other external clock feeds would be a step further back than what is already normal for other products.

Also, you refer just to the galvanic isolation of the fiber connection.  Remember, the EtherREGEN provides true galvanic isolation of data/power/clock to the copper ports as well! (Any 'A' side to 'B' side crossing of the moat.)  Look again at the photo:

 

8 hours ago, justubes said:

The cchd 575 have excellent phase noise, but theres also the non techinal specs of clocks that no one understands affects the sonis largely as well.

Sorry, clocks are science, not magic.  Ultimately low phase-noise at low frequency offsets, and low broadband phase-jitter performance are all that matter for these applications. The circuits of high-performance XOs are already quite low noise from a current draw standpoint, and decent frequency stability is a given.

 

8 hours ago, justubes said:

The reason is if the input master clock can be specified when  ordering? This will allow also the use of say the Mutec mc3+ reclocker for those who have that piece to utilise together with the etherregen.

Sorry, we will not be taking special orders for the external clock input to be set to other than 10MHz.  It does not make any sense to do so.  While the Mutec MC3+ is a fine product, it is not at the level of their REF10, and in fact the MC3+ has its own 10MHz input just for such. Mutec has never published phase-noise/jitter performance specs for the MC3+, and when you consider that and the cable I truly doubt the input of that to the EtherREGEN would be any better than our own internal "femto" clock.

 

Again, this emphasis on the clocks risks overshadowing more important aspects of the EtherREGEN's design.  So as interesting as the topic is, I ask that we move on.  Thanks. 

 

[Gnnett]

Hi guys

 

I was wondering about proximity to other equipment and any associated effects on the system.

 

I am going to use the etherRegen with an UltraRendu and expected to have them, if not on top of, then very close to one another. In the interests of keeping cables short, I intended to connect uRendu to USB dac using Curious regenlink (200mm), usbRegen and the adaptor that came with the usbRegen.

 

So, etherRegen, uRendu, usbRegen and dac all very close together. Hopefully not too silly a question, but if I need to provide greater separation, where do you think it would be best to provide this?

 

Best wishes

 

Grant

 

 

[Superdad]

17 minutes ago, Gnnett said:

I was wondering about proximity to other equipment and any associated effects on the system.

 

I am going to use the etherRegen with an UltraRendu and expected to have them, if not on top of, then very close to one another. In the interests of keeping cables short, I intended to connect uRendu to USB dac using Curious regenlink (200mm), usbRegen and the adaptor that came with the usbRegen.

 

So, etherRegen, uRendu, usbRegen and dac all very close together. Hopefully not too silly a question, but if I need to provide greater separation, where do you think it would be best to provide this?

 

Hi Grant:

Although I am not too sure how USB A>B adapter you mentioned fits into your chain (Curious 200mm Link cable is A>B, already correct from ultraRendu to DAC), as far as positioning I would suggest side-by-side.  That would be better heat-wise for both the ultraRendu and the EtherREGEN.  Not much--if any--in the way of EMI emissions will be coming out of either of those mostly sealed units.

[Gnnett]

Cool.

Curious was from uRendu to usbRegen and A>B was from usbRegen to dac. I think you are saying that output from uRendu no longer requires usbRegen?

Cheers

Gn

[Superdad]

8 minutes ago, Gnnett said:

Curious was from uRendu to usbRegen and A>B was from usbRegen to dac. I think you are saying that output from uRendu no longer requires usbRegen?

 

The ultraRendu and original USB REGEN use the same USB2.0 hub-chip model for output, so it is redundant.  ultraRendu uses the much better--Crystek CCHD-575--clock than what the USB REGEN uses.

 

Different story with ISO REGEN as it uses a USB 3.1 hub chip with better signal integrity, the Crystek 575, five LT3042 regulators, and provides full galvanic isolation via the expensive Silanna ICE08USB.  Quite a few people enjoy the ISO REGEN after their ultraRendus and other renderers, NUCs, and various endpoints.

[Gnnett]

Ah always the salesman eh. 🤣

Yes it is on my radar and the usbRegen was used so that my old USB dac could be powered by a separate LPS.😊

My new USB dac has it's own power supply.

Great news.

Cheers

Gn

[marce]

17 hours ago, Superdad said:

 

I think you are overthinking this.  Please refer to what I has posted previously.  Ours is the only switch to offer isolated dual-data/power/clock domains anyway, so it is not as if potential defeating of the isolation by introduction of "grounds" common to other external clock feeds would be a step further back than what is already normal for other products.

Also, you refer just to the galvanic isolation of the fiber connection.  Remember, the EtherREGEN provides true galvanic isolation of data/power/clock to the copper ports as well! (Any 'A' side to 'B' side crossing of the moat.)  Look again at the photo:

 

Sorry, clocks are science, not magic.  Ultimately low phase-noise at low frequency offsets, and low broadband phase-jitter performance are all that matter for these applications. The circuits of high-performance XOs are already quite low noise from a current draw standpoint, and decent frequency stability is a given.

 

Sorry, we will not be taking special orders for the external clock input to be set to other than 10MHz.  It does not make any sense to do so.  While the Mutec MC3+ is a fine product, it is not at the level of their REF10, and in fact the MC3+ has its own 10MHz input just for such. Mutec has never published phase-noise/jitter performance specs for the MC3+, and when you consider that and the cable I truly doubt the input of that to the EtherREGEN would be any better than our own internal "femto" clock.

 

Again, this emphasis on the clocks risks overshadowing more important aspects of the EtherREGEN's design.  So as interesting as the topic is, I ask that we move on.  Thanks. 

 

Hmmm how do you populate the board with so many chips devoid of text or pin 1 markers?😀

I agree on clocks, did a 64 channel phase array steerable wi-fi antenna, 64 channels with a master clock to sync them all, even on the same board, with total control it was fun! One of the first things as a young PCB designer you are taught though is keep clocks short, at the end of the day on any digital system they are the most important signal...

[Foggie]

On 7/22/2019 at 11:18 AM, marce said:

External clocks are not the best idea for domestic audio, the extra length of cable causes more issues than it solves.

Can you expound a bit on the cable length, as far as fiber is concerned (if it’s any issue)?  

 

I ran fiber to listening room (60ft) and to an area upstairs (100ft) using the two sfp ports in my managed switch in laundry room.

[octaviars]

57 minutes ago, Foggie said:

Can you expound a bit on the cable length, as far as fiber is concerned (if it’s any issue)?  

 

 

The external clock dont have anything to do with cable length of a fiber run they can be many miles long without any problem.

 

The clock signal from a master clock that often is a hi frequenzy signal in most audio cases a 10MHz signal that has to pass through lots of connectors and cables will fast be degraded and not so exact as it has to be. Some DAC manufacturers would never consider building something with a external clock for example MSB Tech or EMM Labs and many more (my own TAD D1000 mk2 DAC has a very precise clock mounted directly besides the DAC chip).

 

This is Ed Meitner from EMM Labs thoughts about external clocks (article in Stereophile magazine).

 

JVS: You don't believe in using an external word clock. Why?

EM: Because I think this is the most stupid thing I've ever heard in the audio business. That means you have a precision clock that you have to connect to a wire to connect to a DAC, when the clock should be straight away where it belongs, inside the DAC, beside the DAC chip, if there is such a thing—not through a cable in a different box. This is so idiotic, it's not even funny. It's a money grab.

JVS: I believe one of the arguments for an external clock is that it offers better shielding from power supplies, and will thus operate better.

EM: Yeah, especially with units [that] have a separate power supply. There's a lot of shielding there. The whole thing is funny.

MP: In our DACs, the reference clock is inside the product and an inherent part of the conversion circuitry. Anything that comes from the outside world, on whatever input we select, is buffered and reclocked, so we don't have to worry about the artifacts of an external clock. Everything is reclocked to our internal reference. That's what we mean by asynchronous clock design.

 

[marce]

3 hours ago, Foggie said:

Can you expound a bit on the cable length, as far as fiber is concerned (if it’s any issue)?  

 

I ran fiber to listening room (60ft) and to an area upstairs (100ft) using the two sfp ports in my managed switch in laundry room.

Nicely answered above, I was going to say similar, the DAC clock needs to be close, the DAC clocks supply if you want to get all microwavey should be under the oscillator so it only has to pass through the board, minimal interference. Any distributed clock should be LVDS. 

[Superdad]

10 minutes ago, marce said:

Any distributed clock should be LVDS. 

 

While you are absolutely correct, almost nobody in audio does so, and we often see DACs with many inches of single-end clock lines (gag!). I do not know why not other than that the clock input of many chips is single-ended.

 

But ALL clock lines on the EtherREGEN are differential (LVDS), and for the parts that take SE clocks we place high-quality LVDS>SE clock buffers within a few millimeters of the chips they feed.  There are 4 on the board. 

And as mentioned before, EtherREGEN is a 6-layer board--which gives a lot of extra control over ground and power planes and impedance control.

[Foggie]

@octaviars + @marce -  thank you and I did read that Ed Meitner article when it published.  However and not necessarily related to the clock "debate" @marce was there something specific to cable length you were referring to above?  And if so is that more for ethernet and not fiber in audio setting as it were.

[Superdad]

1 hour ago, Foggie said:

However and not necessarily related to the clock "debate" @marce was there something specific to cable length you were referring to above?  And if so is that more for ethernet and not fiber in audio setting as it were.

No, I believe Marce was just referring to clock line length.  

Fibre optic and copper Ethernet cable lengths are a different matter--for a different thread.  Would like to keep this one on the topic of the EtherREGEN, thanks. 

[Indydan]

On 7/20/2019 at 1:04 PM, Superdad said:

 

No, both sides are equally “clean.”  As long as you cross the moat—and ideally have the endpoint alone by itself on one side (or the other)—then the EtherREGEN’s isolation and clocking benefits will be fully realized.

 

We are limited to running 100Mbps through the active, low-jitter, differential isolators whch sit on the moat. Since most user Ethernet audio endpoints have little need for data above that rate (even DSD512 is less than half that)—and they may run electrically quieter internally at 100Mbps as Gigabit mode is more current-hungry—it makes the most sense for us to suggest that people use that lone ‘B’ side port as the downstream “output” side.

But as explained (here and in my post just above), there is nothing inherently better about going in one direction or the other. Ethernet is bi-directional signaling anyway! 

Could the lone side be the one attached to the router, and the other side outputs to a Streamer, Apple TV and a NAS?

[Superdad]

19 minutes ago, Indydan said:

Could the lone side be the one attached to the router, and the other side outputs to a Streamer, Apple TV and a NAS?

 

Sure, that will work fine.  But if your "streamer" is attached to the DAC of your best audio system, then we also suggest that you try attaching it to the lone 'B'-side port while attaching your router, NAS, and AppleTV to the 'A'-side ports.  You can experiment and see hear which produces the best results un your system.