External reference for Vivaldi Clock

Hi Greg:
Sorry to have insulted you. Certainly not my intention! And thank you for your purchase of an EtherREGEN.
The EtherREGEN is not an ordinary switch, and its goals, methods, and the ideas underpinning it seem to attract both detractors and misinformation. My response was prompted by incorrect notions about the clocking system of the EtherREGEN.

To be clear:
a) Like virtually every computer/digital product, an Ethernet switch requires a clock. This is of course a data clock, entirely unrelated to DAC audio clock frequencies.

b) For various reasons (explained in our paper) we have determined that both the quality and the handling of this clocking is critical. We go to extraordinary lengths with clocking in the EtherREGEN, using techniques (such as LVDS clock lines and expensive, 10GHz-capable ultra-low-jitter flip-flops) that even surpass what is in most DACs (though I hope some of the upper-end dCS units have something similar).

c) Because we need 4 clock lines (two at 25MHz and two at 250MHz—for the logic chips and flops on both sides of the differential isolation “moat”), we use a jitter-attenuating Silicon Labs clock synthesizer to produce those four LVDS clock lines. Our normal reference for that synth is an ultra-low-phase noise Crystek CCHD-575 that we order custom at 25MHz. Short of a $400 OCXO (cheap OCXOs don’t equal the phase-noise/jitter performance of the $10 Crystek 575) that’s the best production XO to be had.

d) The Silicon Labs clock synthesizer—sitting just millimeters from the most critical logic chips—has 4 outputs and 2 inputs (inputs being for clock it is referenced to). We program the clock synth in house (its outputs, its inputs, and other parameters). We connect the Crystek 575–itself just 2mm from the synth—to one of its inputs.

e) Given that some audiophiles like to play with external reference clocks, and that a near 3x our price competitor’s switch can be ordered with an external clock input (for an extra $250), we thought why not spend an extra $3 on a BNC jack and offer that second synth input for optional use with fancy 10Mhz reference clocks.

It is not something we promote as required, but some people are using it and reporting modest further SQ gain (nowhere near the benefit of just the basic EtherREGEN with its already excellent internal reference clock). [Similarly with power supplies: We spend a TON on the internal power networks, and our switch sounds great with just the custom SMPS we include. But offer a device that runs from external DC and enthusiasts are going to end up experimenting and reporting about it with various fancy linear supplies. And that’s fine. We happen to produce a couple models of highly regarded LPS. But I don’t promote them for use with the EtherREGEN because to me great power supplies are more important elsewhere in a system.]

I am still rather confused about where the comments about our saying anything regards “synchronizing clocking with the DAC” come from.
There is no direct relationship between the clock in the switch and the clock in the DAC (or computer, etc.). There are however effects of ground-plane noise and phase-noise on the ground-plane of an Ethernet input receiver (in the dCS world your Ethernet-input DACs), and those somehow (see our paper) do cause enough disturbance to affect the 0-volt plane by the clock pin of the DAC chip. And that is why people hear a difference with our device (and similar).

I’ve definitely gone on long enough here. I am not the engineer here (John Swenson is) and don’t have all the answers. And we have been upfront that the EtherREGEN was designed based on theories about the underlying mechanisms of all this and John’s vision of how to best directly solve the matter with new chips and a unique and costly design. He continues to work on assembling the highly sensitive test gear needed to produce visual/graphical proof (both for the skeptics and for future development work). We just bought a $15K Jackson Lab PhaseStation and he is building special boxes and power supplies to support the test rig as environmental disturbance adds distracting turmoil to the measurements.

Thanks for reading.
Goodnight,
-Alex C.

1 Like

Thanks for jumping into this thread Alex, gives me an opportunity to debunk the EtherREGEN directly rather than “behind Uptone’s back” so to speak :slightly_smiling_face:

Disregarding the “Optional clock-input” feature, in Uptone’s white paper John essentially outline two broad issues that the EtherREGEN addresses;

(1) Clock Jitter - the implication being that binary digits flowing through the switch might be mis-interpreted if not clocked properly.

(2) Ground-plane noise carried across the Ethernet port into the DAC, caused by (a) clock-jitter causing phase-noise on the ground plane, and (b) leakage current from power supplies causing AC noise leakage into the Ethernet port.

While there is technical merit to the issue of parasitic AC noise traversing the Ethernet magnetic transformers, it’s quite trivial to prove objectively that any standard off-the-shelf Ethernet switch with no special internal clocking and no special noise-blocking will reproduce totally error-free data streams right to the application end-point without any detriment whatsoever.

An Ethernet switch doesn’t distinguish between bitstreams for music versus Compute applications. Error-free data transmissions occurs across millions of networks Worldwide on a daily basis over standard Ethernet switches, completely debunks all of Upton’s other assertions.

Additionally, everything Uptone addresses in the white paper is only at the physical and Ethernet link layers. In actual fact, music streaming involves the TCP layer above that which provides data stream isolation from any physical/link-layer issues. In the case of Qobuz or Tidal streams for example, these TCP flows have significant packet buffering amongst the network elements involved in the end-to-end delivery. Buffers that add up to 10s-to-100s of milliseconds, and enables the incoming packets at the end-point to be error-corrected, re-assembled, and unpacked into the PCM data that can then be fed to the DAC.

Uptone’s white paper is suggesting that the normal clock-jitter and ground-plane noise that occurs on standard off-the-shelf Ethernet switches have an impact on the end-point TCP payload - that’s complete fantasy which only networking neophytes would believe.

Regardless of whether I fully understand the architecture of the EtherREGEN, the fact remains it cannot reproduces data streams to end-points any better (or worse) than a standard off-the-shelf Ethernet switch. A very simple Ethernet BER/Loopback and iPerf Test would readily trash any assertions otherwise within the digital domain.

And any assertions that EtherREGEN’s impact is in the analog domain has already been debunked; https://www.audiosciencereview.com/forum/index.php?threads/uptone-audio-etherregen-switch-review.10232/

Professional Disclosure: I have an Engineering and Physics background and I work for one of the world’s largest Router/Ethernet Switch manufacturer.

I’m closing this thread as it has devolved from a question about the applicability of a 10MHz clock to a debate over UpTone’s claims regarding one of their products. If you would like to discuss one of UpTone’s products the please do so in a more appropriate venue (like UpTone’s forum).

3 Likes