Optical Network Configurations

[Jud]

1 hour ago, jabbr said:

 

The end-to-end testing mandated by the 10G specifications includes testing at the receiver so at the computer if the receiver is SoC, or at the NIC if the receiver is a NIC. I've not seen a shred of actual measurements comparing noise at the board level between 100m and 1g or 100g for that matter.

 

At 100Gbe you could easily transmit the entire song in the short gap between tracks so the network wouldn't be an issue!

 

🙂 There's a thought!

 

1 hour ago, jabbr said:

 

That said it is reasonable to use a low powered, low noise endpoint like the Fitlet or the Clearfog or the OpticalRendu ... I think its very reasonable for something like the OpticalRendu to be designed for low noise and to optimize the USB output. My own impression is that there may be factors on the board that affect SQ via the USB output and use LPS etc for my own endpoints. 

 

The discussions I've had regarding blocking upstream factors via 10Gbe+ e.g. servers, NAS have to do with the mandated stressed receiver testing, and see no reason why downconverting the signal to the endpoint should be an issue -- this is what I've been doing -- my low powered ARM SBC 1g endpoints sound just as good as the 10g NAA that I had built using a Celeron board. The sensitivity of DACs to USB output also is DAC dependent to my ears.

 

Agreed about the low noise endpoints/power. I haven't listened to enough DACs to have an opinion about the sensitivity of various DACs to what's coming in to the USB input, but the strategy of using low noise endpoints should help if a DAC is sensitive. I also use an isolator. Miska (presumably based on measurements) says the Intona works, so I'm supposing the general notion makes sense.

 

1 hour ago, jabbr said:

 

Does the OpticalRendu do 100m Ethernet via fiberoptic?

 

Don't know, have the microRendu (no optical input) myself.

[jabbr]

6 minutes ago, Jud said:

Agreed about the low noise endpoints/power. I haven't listened to enough DACs to have an opinion about the sensitivity of various DACs to what's coming in to the USB input, but the strategy of using low noise endpoints should help if a DAC is sensitive. I also use an isolator. Miska (presumably based on measurements) says the Intona works, so I'm supposing the general notion makes sense.

 

The notion of an isolator being "complete" and the upstream jitter, noise etc getting through to the DAC is an oxymoron to me.  I'm willing to consider that USB isolators are only partially effective. @Miska I believe also has stated that the Holo May DAC is not improved by USB isolators. The fact that "audiophile" designed devices are routinely used with other isolators befuddles me. The trickiest noise to isolate is common mode which is why I started down the fiber path to start. I have 100% confidence that common mode noise is not being passed down my fiber cable, more confident than with any electrical moat.

[Jud]

1 minute ago, jabbr said:

 

The notion of an isolator being "complete" and the upstream jitter, noise etc getting through to the DAC is an oxymoron to me.  I'm willing to consider that USB isolators are only partially effective. @Miska I believe also has stated that the Holo May DAC is not improved by USB isolators. The fact that "audiophile" designed devices are routinely used with other isolators befuddles me. The trickiest noise to isolate is common mode which is why I started down the fiber path to start. I have 100% confidence that common mode noise is not being passed down my fiber cable, more confident than with any electrical moat.

 

Yes, it's not the fiber I'm trying to do anything with, it's the USB connection from endpoint to DAC. (My DAC is considerably less expensive than a Holo May, so I don't know if that example is strictly relevant. 🙂)  I think I hear a difference, but happily accept that I could well be imagining things.

[jabbr]

54 minutes ago, Jud said:

 

Yes, it's not the fiber I'm trying to do anything with, it's the USB connection from endpoint to DAC. (My DAC is considerably less expensive than a Holo May, so I don't know if that example is strictly relevant. 🙂)  I think I hear a difference, but happily accept that I could well be imagining things.

 

 I'm saying that when I use the term isolation, it means that upstream stuff doesn't get through. The term "Rejection ratio" is used to indicate the degree to which a system rejects noise from getting through. "PSRR" is the degree to which a power supply rejects noise from input. "CMRR" describes the degree to which common mode noise is rejected -- pretty darn high for fiber!. "PNRR" could be "phase noise rejection ratio" and if the range of 1g and 10g switches were compared, we would see that the "PNRR" of the newer switches would be much greater (as specified). It is common for components to be specified this way on their spec sheets.

 

When someone tells me that upstream noise gets through, or that an upstream cable has an effect on SQ, it means to be that the rejection ratio of their isolating devices is less than the minimum inaudible dB.

[Jud]

1 hour ago, jabbr said:

 

When someone tells me that upstream noise gets through, or that an upstream cable has an effect on SQ, it means to me that the rejection ratio of their isolating devices is less than the minimum inaudible dB.

 

Sounds better to me with the USB isolator. That is all I know; whether, to paraphrase Keats, it is all I need to know, well, I don't know. 😉

[Superdad]

On 4/18/2023 at 5:55 AM, Jud said:

For example, I recall the reason it was said the ER had a 100M interface was because at the time there weren't the associated 1G components with the electrical noise characteristics (and presumably price) needed.

 

Not quite. Had nothing at all to do with "electrical noise characteristics." 

Allow me to explain...

EtherREGEN (Gen1) has four gigabit-capable RJ45s and one gigabit-only SFP on its 'A' side and a lone 100Mbit-only RJ45 across its active differential isolation 'moat' on the 'B' side. The reasons for the 'B' port being 100Mbit-only were manifold:

a) The fastest low-jitter digital differential isolators available at the time were 600Mbps--and those are not fast enough for Gigabit Ethernet data rates;

 

b) We used the T.I. transceiver on the 'B' side in 100Base-TX "fiber" mode (to 100Base-T copper) and that is the format we put across the moat. Even if the faster Gigabit isolators (which became available and which we sometimes substituted because of better stock) were available, it was not possible to have the T.I. part--in that fiber>copper mode--do proper auto-negotiation with the connected device and back across the moat to the main switch chip. Besides, at least half of the endpoint devices at the time were 100Mbps, so being Gigabit-only was not an option.

 

c) A key aspect to the EtherREGEN's unique design is our use of 10-gigabit-capable ultra-low-jitter flip-flops for reclocking the signal on both sides of the moat. It is important because even the very best digital isolators add a small amount of jitter, and reducing such is a goal of our design (this is because both common-mode leakage and jitter/phase-noise are culprits in propagating chip-and-ground-plane noise downstream). We use two of these these rather special flip-flops--and the buggers are now up to $20 each!

Anyway, when reclocking with a flip-flop, you need to feed it a clock that is at least twice the data rate of the signal you are reclocking. (My non-engineer brain always forgets the details of this, but I think the flop can work off both the leading and trailing edge of the clock, so the clock does not have to be 2x.)

So for EtherREGEN the main clock synthesizer on the 'B' side produces 4 differential clock outputs: two 25.0MHz clocks (one for the 'A' side main switch chip and the 'A' side's T.I. transceiver, and one sent across the moat to the T.I. transceiver/PHY handling the fiber-mode>copper duty); two 250MHz clocks--one for the 'B' side's flip-flop, one for the 'A' side's flip-flop. That 250MHz clock gets fed back to the 'A' side through our second differential isolator.

If we changed the format of what we put across the "moat" and used the now available Gigabit-capable isolators, with the original topology we would have to put a 1.25GHz clock through an isolator.  But that's faster than any available isolators!

 

However, for EtherREGEN Gen2 we are moving to auto-negotiating Gigabit-capability. We have changed the topology and will now be putting SGMII across the moat, with just a brand new ultra-low-power SGMII>copper PHY (hard to find!) handling the 'B' port with 10/100/1000Base-T.  (The new main switch chip is already tested/implemented, and the 2+ months overdue new-model PHY samples are supposed to finally ship to @JohnSwenson this week; The bare board for this next development stage has been ready and waiting for these chips since early February.) 

To accomplish this we are of course now using the relatively more available Gigabit isolators (still not great supply out there of anything, hence I have been stockpiling parts for over a year--about a $100K of chips sitting in the corner!). And yes we have to synthesize 1.25GHz for the flip-flops, so we are moving away from SiliconLabs/Skyworks and to the Analog Devices ADF series--which have even lower jitter.  But since it is still impossible to put that clock rate over an isolator, we are only using our special flip-flop on the 'B' side.  So while EtherREGEN Gen1 was performance-wise pretty symmetrical (B>A was almost as good as A>B--which we did primarily to make the SFP cage a really good output to optical endpoints) that will not be the case with EtherREGEN Gen2. The 'B' port will likely be markedly better.  [Not that the 'A' side will have anything to apologize for: Even just the 'A' side is a terrific well-clocked, well-isolated switch with a pile of expensive LT-3045 regulators and 12-core-per-port magnetics.] 

 

 

On 4/18/2023 at 8:26 AM, jabbr said:

Does the OpticalRendu do 100m Ethernet via fiberoptic?

No, the opticalRendu's SFP input is Gigabit.

[Jud]

1 hour ago, Superdad said:

a) The fastest low-jitter digital differential isolators available at the time were 600Mbps--and those are not fast enough for Gigabit Ethernet data rates;

 

The availability of low-jitter isolators is what I inaccurately remembered as availability of low noise parts.

[jabbr]

@Superdad thanks for the explanation. 

[SQFIRST]

On 4/18/2023 at 2:40 PM, jabbr said:

The term "Rejection ratio" is used to indicate the degree to which a system rejects noise from getting through. "PSRR" is the degree to which a power supply rejects noise from input. "CMRR" describes the degree to which common mode noise is rejected

@jabbr Thank you for your helpful distillations throughout the thread. The nuggets such as the terms above are good to be aware of while trying things out. This thread has been of great help in learning about optical networking and isolation.

[audiophilac]

for eR Gen2, would there be any difference on SFP Transceiver heat?

[Superdad]

6 hours ago, audiophilac said:

for eR Gen2, would there be any difference on SFP Transceiver heat?

Not at all.  The current draw—and heat—of an SFP transceiver is determined by the transceiver not by the cage it is plugged into.

 

That being said, we do dedicate an LT3045 regulator to the SFP provide the 3.3V required by the transceiver, and that does generate a little bit of heat depending upon current draw. But since all dozen of the LT3045s employed in our switch are only ever fed 0.7V above their set output voltage the drop is only that—so dissipated heat per reg is only 0.7*I = W (‘I’ being current drawn and ‘W’ being wattage dissipated).

[Reg19]

Forgive me: I got a headache when I tried to read the last 10 pages as I’m not that conversant / knowledgeable on this topic.🙏

 

 

 

I do own two streamers: (1) SoTM Trifecta with SPS-500 and SBooster as power supplies, and (2) opticalRendu + opticalModule with two Sonore Audiophile Power Supplies with some generic TL-SM311LM SFP modules.

I also have the Innuos Phoenix Net switch (for trial) at home that I’ve connected my NUC (running Roon), Mac Studio M1 (running HQPlayer Desktop) and streamer. My DAC is the Holo May.

 

Currently, I prefer the SoTM setup over the Sonore one (it has better timbre and seems more ‘natural ‘ organic’. The latter seems a bit ‘digital-edged’ to me with a flatter soundstage (albeit, having perhaps quieter background).

 

So, my question is:
1. Can I improve the SQ from my optical setup (#2) by changing the SFP modules? If so, which one would you recommend?

 

Thanks, in advance for helping me out. 

[jabbr]

https://www.servethehome.com/qnap-qsw-m7308r-4x-is-a-4x-100gbe-8x-25gbe-switch-marvell/

[Jud]

Have installed a Fitlet3 with optional SFP+ input. (It works fine with 10G transceivers on both ends of the optical fiber feeding it, though it doesn’t provide speeds higher than 1.25G.) So everything from the remote computer to the UPnP endpoint/NAA (that is, the Fitlet3) is optical Ethernet with SFP+ connections.

 

From there, it’s USB through an IsoRegen to the DAC.

 

 

 

 

[jabbr]

Time Cards with Atomic Clocks at OCP Regional Summit 2023 Prague (servethehome.com)

[Jud]

1 hour ago, jabbr said:

Time Cards with Atomic Clocks at OCP Regional Summit 2023 Prague (servethehome.com)

You’re thinking about this as an ultra-low-jitter clock ultimately available to consumers?

[fas42]

The stability of the waveform in an optical link can be a key determinant of the SQ; the ideal is to have the receiving area of the chain completely isolated, or not vulnerable, to variability of the signal - my current rig which relies on a low cost DVD player to generate this link data at this stage of its tweaking demonstrates this behaviour, virtually every time a CD is loaded: that is, the SQ alters depending on how 'comfortable' the player is with the seating of the disk. Of course, if the chain following the optical link was completely impervious to the qualities of the waveform, fully robust, then it wouldn't matter ... but, the real world intervenes .

 

Can a value for money solution be devised, at a cost commensurate with the other equipment? I'm finally in the process of attempting a trial hardware solution to this, having obtained the key part - and starting to play with it. Will it do the job? Conceptually, yes ... my Edifying Journey thread will detail how things progress ...

[jabbr]

On 7/17/2023 at 3:14 PM, Jud said:

You’re thinking about this as an ultra-low-jitter clock ultimately available to consumers?

Eh really the best low jitter clocks are built into the equipment itself. External clocks have higher jitter, thats just physics. The application for external clocks is when its necessary to **synchronize** multiple pieces of equipment to the same time ie PTP. As we get into multichannel with distributed DACs is when it's important for this synchrony as is specified in Ravenna etc.

[audiophilac]

On 2/4/2022 at 1:55 AM, audiom3 said:

I started off with the galvanic isolation method of utilizing dual FMCs right before the streamer/bridge.  Then I did the same for media streamers (Nvidia ShieldTV and Roku Ultra).  So surprised by the improvements, I just couldn't stop.  Eventually changing over my entire LAN from copper to fiber.  Currently, my setup looks like this:

I have 1G fiber into my home which utilizes their own proprietary fiber converter.  I use a very short ethernet cable out of it and into my ASUS RT-AX89X router's WAN input.  At this stage, the only points in my network that employ copper again are just before each video streamer, my PS Audio Bridge II and a short run out of my Intel NUC.  I run fiber (OM4) out of the router using the 10G SFP+ cage and into my first Mikrotik CRS309 switch.  This way my local playback doesn't have to go through the router at all.  My NAS, another Mikrotik CRS309 and Intel NUC are the connected devices to this first switch (via OM4 fiber).  The exception being my NUC, which uses a pair of Finisar FTFL1318P3BTL transceivers and OS2 fiber into a Sonore optical Module (OG model).  I have a 50' run of OM4 to my 2nd CRS309 switch which is in the front of my media room behind all of my components.  Off of this switch is a short run of OS2 fiber into an EtherRegen (also using the same Finisar transceivers as above) then into my PS Audio DS DAC via a WW Starlight 8 ethernet cable.  Various video streamers are fed from this switch and they all utilize linear PSUs for their FMCs.  Every switch, the router, the Sonore oM, ER and my NUC run on linear PSUs.  My server houses my HDDs for movies and SSDs for music, so I didn't bother with an HDPlex/ATX linear setup for it but it does utilize an Intel X520 NIC (SFP+ NIC).  This way it never has to convert copper to fiber externally.  I may try an HDPlex in it someday just to hear the difference.  I am currently in the process of switching out all SFP/SFP+ from generic (mostly 10Gtek) to Finisar modules.  I don't know what it is about those Finisars but they improve every device I put them onto. 

This whole 'project' turned into a full-time hobby but I've been thoroughly enjoying it and learning so much.  I enjoy reading about new things to try and then how to implement them.

Hi,

Is your CRS309 still connected to etherRegen?

[audiom3]

3 hours ago, audiophilac said:

Hi,

Is your CRS309 still connected to etherRegen?

I sold the ER after I purchased a Lumin U2. With it's optical input, the ER became obsolete. 

[Superdad]

4 hours ago, audiom3 said:

I sold the ER after I purchased a Lumin U2. With it's optical input, the ER became obsolete. 

I guess more redundant for you than “obsolete.” 

You have an optical network feed from you Microtik.  
Yet there are a good many Lumin X1 and U2 owners feeding optical from their EtherREGEN. We went to a lot of trouble and expense to make the EtherREGEN’s performance—data/clocking/power, but not port compliment—symmetrical on both sides of its active differential isolation moat, so that it could be used in the B>A direction with fine results.  We did so specifically for optical endpoint users (which at the time was only the Sonore opticalRendu, Lumin models subsequently arrived on the market).

 

Separately, @audiophilac just sent me a PM regarding some troubles feeding his EtherREGEN from a Microtik 10G switch.  To him I advise two things:

a) Use SFP transceivers and not SFP+ to assure that the Microtik stays at 1G;

b) Read/search this thread for reports from other Microtik users as I recall someone here saying that the Microtik settings to force a port into 1G do not actually work. There seems to be some confusion/contention around this.

[audiom3]

3 hours ago, Superdad said:

I guess more redundant for you than “obsolete.” 

You have an optical network feed from you Microtik.  
Yet there are a good many Lumin X1 and U2 owners feeding optical from their EtherREGEN. We went to a lot of trouble and expense to make the EtherREGEN’s performance—data/clocking/power, but not port compliment—symmetrical on both sides of its active differential isolation moat, so that it could be used in the B>A direction with fine results.  We did so specifically for optical endpoint users (which at the time was only the Sonore opticalRendu, Lumin models subsequently arrived on the market).

 

Separately, @audiophilac just sent me a PM regarding some troubles feeding his EtherREGEN from a Microtik 10G switch.  To him I advise two things:

a) Use SFP transceivers and not SFP+ to assure that the Microtik stays at 1G;

b) Read/search this thread for reports from other Microtik users as I recall someone here saying that the Microtik settings to force a port into 1G do not actually work. There seems to be some confusion/contention around this.

Yes obsolete. My entire LAN is fiber optic, so it would be foolish to convert from light to copper and then back to light just to utilize an ER.

[R1200CL]

11 hours ago, Superdad said:

b) Read/search this thread for reports from other Microtik users as I recall someone here saying that the Microtik settings to force a port into 1G do not actually work. There seems to be some confusion/contention around this.

This is Microtik answer:

 

Hello,

The CSS610 do not support setting Rate Select (RS0) signal to the Low state.
Unfortunately, we cannot make 1G support in CSS610 for 1G/10G SFP+ modules which firmware strictly follows that.
All MikroTik SFP+ modules also support 1G/10G speeds and work in both speeds in the CSS610, we can only suggest using them instead.

Sorry for inconvenience.

Regards,
Jānis B.

[R1200CL]

13 hours ago, audiom3 said:

My entire LAN is fiber optic, so it would be foolish to convert from light to copper and then back to light just to utilize an ER.

The ER should also be looked as a DDC. If the ER2 will have fiber both sides, it may be worth trying the ER2. Especially with an 50 ohm external 10mHz sine wave clock.

I’m using an opticalRendu, and I do the “foolish” conversion with a 10 GB Mikrotik upfront 😀

 

[audiophilac]

Would there be another Layer 2 SFP+ switch, fanless that could work with eR?

I'm currently using Aruba POE switch with four SFP+ ports.