The EtherREGEN thread for various network, cable, power experiences and experiments

[dbastin]

51 minutes ago, barrows said:

every device must re-clock, asynchronously, along the way, or else the accumulated timing errors would become so large that data would never make it intact

Thanks, pretty much everything you said is my understanding however that last statement as quoted might not be true for optical in - optical out devices according to what John stated.  This is why I have queries, to clarify.

 

Internet at my house comes via GPON fibre network of the national broadband network. 

But if fibre gear does not clock out then I am receiving transmissions that are passing on the result of the last wired device ... and who knows how well that is deciphered by my devices.

 

For interest, PON is Passive Optucal Network, meaning things that split or agglomerate fibre are passive devices ... no clocks. I'd guess the exchange has active devices like switches, but may not clock optical outputs.

 

The nearest exchange used to be 300m away and internet was pretty much faultless and fast.  That closed and now the exchange is 10x further away ... now there is more often sluggish times and frequent drop outs. Maybe that is pnly because the new exchange serves alot more users.  Maybe more devices between the exchange and me.

[dbastin]

8 hours ago, audiobomber said:

Reclocking is done in the FMC

But John said ...

On 2/29/2024 at 4:06 PM, JohnSwenson said:

The optical connections do NOT make the timing perfect. They do NOT "reclock" the data. whatever jitter that is on the input is still there on the output

 

8 hours ago, audiobomber said:

I went from wi-fi to copper to optical, and each step was a sonic upgrade. 

If fibre does not clock out, then wifi could be superior if it clocks out and because it also blocks common mode noise.  But that depends if the wifi devices are not inherently doing more damage in other ways.

 

My principles include;

• Use fibre for longest runs (due to price : performance ratio)
• Use cable for shortest runs when there is no option for fibre, because the cables cost so much if their performance is on par with fibre.

Wifi seems to need careful deployment.  I use wifi to connect to my Devialet, it is at least as good as Auralic Aries G1 in terms of SQ over wifi.  And both were superior to wired connection with daisy chain of 2 ERs and a Gigafoil. i will revisit this because many things have changed since that time, including my knowledge.

 

In my experience with Cisco SFPs and comscope SM between ER and Ubiquiti 1G switch, fibre competes with US$1k 'audiophile'  cables of 1m, 10m of those  cables is just not cost effective.  However the cables have sound qualities that could be preferable (that may not be the case with better SFPs, SFP+, AOC).

[dbastin]

Thanks @Superdad Alex for clarifying.

 

In hindsight I gather ....

On 2/29/2024 at 4:06 PM, JohnSwenson said:

The optical connections do NOT make the timing perfect. They do NOT "reclock" the data. whatever jitter that is on the input is still there on the output,

... was referring to SFP modules rather than the entire 'connection'.  Phew, that is what I had understood up to that point.

 

So just to sum up the answers to my questions/comments (for clarity).

 

15 hours ago, dbastin said:

For optical transmission, I think you are saying the transmission is not clocked out so therefore timing variances/errors not corrected just passed on.

That is incorrect, all ethernet is clocked out but not by the SFP module.

15 hours ago, dbastin said:

If transmission to the endpoint/DAC is wifi, is the timing (and jitter) still created by the wifi access point (WAP) clock?

Yes, it will be clocked out by the WAP with or without impections / variances due to the quality of the WAP design/parts.

 

15 hours ago, dbastin said:

What I am unsure about is if a switch with a top performance clock will 'correct' timing errors when sending wired transmission - that is, if the transmission the switch receives has timing variations/errors that those are reduced, minimised, eliminated

I gather the answer is yes, but within the capabilities of the switch design/parts (eg. It may become less imperfect - how much less depends on the switch).

 

15 hours ago, dbastin said:

Finally, your ER moat's 'reclocking':

Does the moat correct incoming timing variances/errors and superimpose its own better timing?

Is the moat in addition to what clocks out the ethernet transmission from Side B? 

What clocks out Side A RJ45 ports?

Does the moat correct incoming timing variances/errors and superimpose its own better timing?

Not really answered.  The whole design probably reduces/minimises errors.

 

Is the moat in addition to what clocks out the ethernet transmission from Side B? 

 

Yes, the 'signal' clocks put via the switch chip and then converted, put through and reclocked by the flip flops, converted back to UTP.

 

What clocks out Side A RJ45 ports?

The switch chip on Side A (the flip flops have bo role here).

 

I have a question.

 

Other than the isolation, does the reclocking across the moat improve the RJ45 output timing? (eg. Further reduce imperfections / variances of the signal coming in on Side A)

 

If this has been answered in the past 122 pages, feel free to point me to it.

 

Cheers

[dbastin]

1 hour ago, Superdad said:

once the data arrives over the isolators it is then in a completely walled off domain--consisting of completely isolated low-impedance power networks, isolated differential clock lines, a single port PHY chip, and most important of all, the super-secret sauce of the EtherREGEN: the second instance (there is one on the 'A' side also) of out 10GHz-capable ultra-low jitter differential reclocking flip-flop--running at 250MHz.

Those last elements--the flip-flop feeding the Texas Instruments PHY--are what produce the pristine final output of the EtherREGEN.

Thank you again, although you've probably written this just a few times in the past.  It is good to reiterate in the context of my observation that:

• the last 'clocking' thing before Side B RJ45 is the flip flops (ie. switch chip > PHY Side A > flip-flops & buffer > RJ45 incl. its magnetics)
• The last 'clocking' thing on Side A is the switch chip (and the flips-flops are not involved - unless the signal has come via Port B)

So compared to Side B, Side A is affected by noise from the SFP and other things plugged into its RJ45 and does not have outputs via flip flops.

 

Noise on Side A can be minimised:

1. I presume (or hope?) there are SFP modules that generate less noise than common mode noise than is injected via a cable in a RJ45.
2. connect the least possible to Side A RJ45 ports 
3. put caps in the unused RJ45, eg:

• https://www.audioquest.com/accessories/audio-enhancements/noise-stopper-caps/rj45-noise-stopper-caps
• https://www.myhifishop.de/Voodoo-Ecke/AQVOX-LAN-Detoxer-RJ45-Kappen-LAN-Kappen-Netzwerk-Entstoerung::79.htm (I find these really good)
• https://quantum-science-audio.com/qsa-lan-jitter
• https://www.synergisticresearch.com/accessories/uef-performance-enhancer/

What I have done (if you didn't notice on Page 121) is put the signal from the internet and router into ER via SFP, then across the moat to the server, and then back across the moat to Side A RJ45 towards the endpoint.  So I am using the flip-flops twice.

 

My feeling is the server benefits from pristine signal, and Side A benefit from being isolated from noise generated by the server (after all, it is a computer, even though my Antipodes CX with Oladra upgrade is really excellent).  I suspect, even with some relatively minor compromises on Side A (as noted above), this configuration is certainly worth considering.  The signal into switch chip from Side B is quite 'pristine' and reclocked by the flip-flops, but may be diminished by the switch chip itself which is also dealing with the signal coming in from the SFP.

 

That is the theory, I have started to compare with listening.  The first session was not serious listening because the cables have been moved and might need to 'settle' after changing to Side A server Side B endpoint - so it was casual listening.  Back to back comparison is not really possible.  After a few hrs I felt the Side B server Side A endpoint config is more like PCM character whereas the SIde A server Side B endpoint config is more like DSD.  They are different, the difference between them is not as large as my previous comparisons of PCM and DSD64 though.  The SIde A server Side B endpoint config has more atmosphere and clarity, high frequencies like chimes are super clear and cut through, but the low frequencies are more towards gentle than ultra-defined (the ultra-defined might be a hard edge caused by distortion).  It will be probably very hard to determine if one is 'better' - the difference is not great and preferences are in the mix. 

 

Incidentally, the audio data is from Tidal via Roon Core as PCM upsampled to 188/192 24 bit and then via Devialet AIR to the 'player/renderer' in Devialet Pro (Core Infinity board).  This arrangement sounds most accurate and real to me (compared to roon core to roon player, and comparing PCM and DSD - I have not tried HQ Player yet).  I will also listen to material off the SSD in the server (HD Tracks downloads) so the switch chip is under less load.

 

I might have to try upsampling to DSD64 with Side A to endpoint to blend the characters.  If I change players, then the blending may be different again.

 

I hope this is of interest.

 

PS. perhaps the next ER (or a deluxe version) could have 2 moats and 3 domains.

3 hours ago, Superdad said:

The differential isolators are what block common-mode noise and leakage from getting to side 'B' (better than any passive transformer/inductor based methods).  

Although I suspect it is the flip flops that really make the difference, I suppose you considered optical methods.  I gather PS Audio has that in its new streamer. 

[dbastin]

On 3/4/2024 at 4:55 AM, Superdad said:

ultimately we are trying above all to benefit the DAC-attached endpoint (or Ethernet-input-equipped DAC) with the best signal. And for a variety of reasons that is generally always going to be the 'B' port.

After 24 hrs of cable etc 'settling' (or whatever) and them another few hrs including a bit of tweaking and listening, I think that Side A server Side B towards endpoint is best ... but of course you will not be surprised.  So extra reclocking doesn't seem superior to Side B.

 

On 3/4/2024 at 4:55 AM, Superdad said:

special reclocking flip-flops.  So both are needed.

Both are needed to get the Side B result, but the point I was making is the flip flops are what sets ER design apart from typical isolation that doesn't.

On 3/4/2024 at 4:55 AM, Superdad said:

Paul McGowan officially confirmed that they are NOT using optical digital isolators in the AirLens

My mistake, I was sure I has heard or seen that they are.  In a Youtube Paul 'vaguely' mentions RF .  He also implies it is extremely rare for a streamer to isolate its output stage - maybe he hasn't seen Auralic, Aqua ...

2 hours ago, JohnSwenson said:

I was there for 33 years and it all started with a flipflop.

That's flipping amazing, you're a flopping genius!