jabbr

I’ve been using Qobuz for purchases for some time now ... I initially had to, err, travel to the Netherlands to sign up, because Cécile McLauren Salvant wasn’t, err available, in the USA ... but I’m, err, back in the States and will seriously consider the Ultimate plan, so I can both listen & download ... call me old fashioned but I like to buy albums that I really like.

Well ... nothing to suggest that 1.5m is better for hypothetical USB reflections in any case ... Shoot

Seems to be Corning Clear Curve. The main benefit for the home user is that it can be bent to a much greater degree without breaking.

Yeah so the FTLX1471D3BCV is the dual 10G/1G version, and the suffix "-IT" means Intel chipped and that's actually what I use. There's also a single mode 1000Base-LX SFP module which is probably a bit less but remember that the single mode optics (VSEL) are more expensive than multi mode. I think this is the 1G part: https://www.mouser.com/datasheet/2/610/finisar_ftlf1318p3btl_industrial_temp_1.25g_rohs_c-934368.pdf I use the dual 10G/1G for my own flexibility, and because although I haven't actually tested it, and the 1G SFP modules work perfectly fine, the 10G parts are designed to work with significantly higher standards. There was a huge jump in testing requirements when the 10G specs (10GBase-X) came out, including stringent end to end jitter etc. Not that 10G single mode is needed in the home environment but its just a few $ more and crazy good Now if you are going to pay >$1000 this is the puppy you want: https://www.mouser.com/ProductDetail/Finisar/FTLC1154RDPL?qs=gt1LBUVyoHlQo6N0RZ1q1Q%3d%3d 100G over a single mode fiber! Imagine the eye diagram that needs to hit Oh and < 3.5 W

I don't spend that for SFPs. I don't have any Foxconn and know nothing about their quality. Finisar (and Avago) are OEMs that actually make the SFPs for Cisco and others (Intel uses both Finisar and Avago). For example the Finisar FTLX1471D3BCL is named FTLX1471D3BCL-CS (or something like that for the Cisco version which has the Cisco code burned in). e.g.: https://www.intel.com/content/www/us/en/support/articles/000005528/network-and-i-o/ethernet-products.html

http://www.ascdi.com/anti-counterfeit/cisco-counterfeit-indicators/ https://packetpushers.net/overpriced-optics-by-oems/

The impedance is specified 90 ohms +/- 10% (ish). Meeting spec doesn't mean that reflections don't matter at all, rather that they are within limits. In my use of language a small number is not the same as zero. In any case the specific question posed to me did not suggest, to me, that the question was limited to the USB specification, which frankly interests me very little, but rather to the issue of reflections in cables which I find more interesting in general, as well as reflections in PCB traces, around vias etc etc etc, which I do find relevant to the topics that I'm interested in. So suit yourself. 👋

Nice analysis I agree that the jitter on USB, given buffering etc, is not nearly the concern as for a DAC clock. My point was simply that the amount of correlated jitter (as well as other sources) is order n of magnitude higher than random jitter for the USB clock itself thus placing a bound on the amount of jitter that ought be tolerated on the USB clock itself -- also this analysis can be similarly made for external clocks and again, small discrepancies in cable impedances can cause small amounts of reflection which result in small amounts of jitter yet still orders of magnitude more than the clock oscillator itself. This point is more appropriate for other recent threads but @Ralf11 asked specifically, and this is a specific answer: yes reflections matter to a very limited degree if the cable meets spec. These same issues also apply to Ethernet and the specific reason that I recommend Belden bonded cables, and particularly the REVConnect connectors, is that they maintain a very constant impedance through the connector, cable and connector also minimizing reflections. In any case there seems to be recent fascination with using "OCXO" clocks in USB, Ethernet, external clocks etc etc etc and regardless of the unconcern with very slight reflections in cables, any clocked signal which travels over a cable will be subject to such types of correlated jitter that use of "extreme" clocks is not warranted.

Hard to know what level of jitter is of concern, but the point should be ala the meme de jure , is that while 1.6% does seem like a small number, such a level of correlated jitter vastly outweighs the types of clock jitter that are being talked about. ie -120 dB is about 1e-6. and that’s for on spec cables not audiophile cable’s! Again, that + crosstalk etc more important than TCXO vs OCXO. Of course some people’s golden ears appear to be able to resolve better than most of our measurement equipment. Perhaps NASA should hire some audiophiles to assist with deep space exploration?

A digital signal should have a constant rise time and within spec to allow the receiver to trigger at the same point/phase. A variable rise time will cause jitter in the receiver. A reflection can alter the signal/rise time between samples and thus cause jitter in the receiver. In a similar fashion crosstalk causes correlated jitter.

All of the Class-D type amplifiers use a feedback loop, analog for Class D or digital for DDFA, but consider a DAC itself: there is no such feedback loop prior to the I-V stage. Our DACs prove that it’s entirely possible to produce a low distortion current amplifier with zero feedback

Yes! This type of design interests me quite a bit. The issue is controlling transmitted EMI from before the filter! Doable but not entirely trivial.

It’s the future. At 45 MHz you can build a wireless transmitter and then the DAC/amplifier/speaker can broadcast your music across the whole town 😂 The only problem is that some aircraft communications systems use that frequency band, otherwise ... nothing a little strategically placed mu-metal can’t fix

Right, so perhaps my attempt at a time domain analysis isn’t helpful. Frequency domain is easier and 1/22e-6 @ > -100 dBc is a very very small number

No here I’m talking about the +/- 1 Hz component which is at -100 dBc/Hz (phase noise level) ie the error signal compared with the center 22 MHz. That error ripple will itself be attenuated by the analog LPF.