Post B6QSVGiOnxOAqlNP1c by Stormgren@obsidianmoon.com
(DIR) More posts by Stormgren@obsidianmoon.com
(DIR) Post #B6QOpPyvf5axZ5ScyG by Stormgren@obsidianmoon.com
0 likes, 0 repeats
I don't think a media converter can't get any more simple than this...Nominally 1G and below convertible, I tested it at 10g over the weekend via iperf3 and generic 10g plugs seem to be okay with it. No active cooling so higher-power plugs might be an issue.Pretty sure the LM358 is used as a link detector of some kind by monitoring the LOS pin and lighting the port LED accordingly.(And yes, I actually need it that simple for some of the current nutty stuff I'm working on at work. Retiming won't be my friend here.)
(DIR) Post #B6QOpQCOr0N2ErlNy4 by azonenberg@ioc.exchange
0 likes, 0 repeats
@Stormgren I have a very similar one I built myself except it doesn't even have any power regulation (takes 3v3 from a bench supply). Just a couple of capacitors and filtering components.(There's no need for the mid-span AC coupling caps here, per SFF-8431 the modules should contain their own DC blocking caps.)My version is designed for 1/10G and I might eventually respin with a bit of layout tweaks and more robust connectors to have better performance at 25G.The key unique attribute of mine is that rather than being 1:1, the TX and RX diffpairs are fed into 6 dB resistive splitters with one half of the path going to the opposite side optic, while the other goes out to a SMA allowing you to do mid-span protocol decoding/sniffing. Great for PCS/MAC level debugging
(DIR) Post #B6QQ6VgjnWum8mfLSi by Stormgren@obsidianmoon.com
0 likes, 0 repeats
@azonenberg I was also thinking that those coupling caps were slightly suspicious. I wonder if it's to break up the trace lengths more than be a blocking capacitor between the plugs?As for your custom breakout, that sounds incredibly handy! I've considered designing something like that myself, but just don't have enough spare time right now. I've been looking for SFP breakout boards for another experiment we're looking to run, and they're either crazy expensive or kinda unobtanium. Best option I've found for a somewhat reasonable price is the Osmocom experimenter boards with the built-in differential drivers, as I need to inject arbitrary single-ended signals into a SFP driver and also be able to read the I2C and other pins as well.If you have any thoughts on breakout board sources, designs or whatnot, I'd really appreciate hearing them.
(DIR) Post #B6QQaHZ6s5fr1hIJrU by azonenberg@ioc.exchange
0 likes, 0 repeats
@Stormgren I'm gonna respin mine mostly because the amphenol edge launch SMAs on it are a pain to assemble and the solder tends to shear when torquing it after a couple of dozen fatigue cycles.And as long as I'm doing that might as well do some other tweaks
(DIR) Post #B6QQcprgJkPVz3CNkG by azonenberg@ioc.exchange
0 likes, 0 repeats
@Stormgren will probably swap for rosenberger 32k243-40ml5 or similar
(DIR) Post #B6QR32LcWrJMEBhDsW by azonenberg@ioc.exchange
0 likes, 0 repeats
@Stormgren eventually i want a whole suite of phy breakout boards available for testing just a matter of having time to do a good job.I also want to design the future ones with 2x thru's for de embedding feedlines etc
(DIR) Post #B6QRGkY5VAL5504BJw by azonenberg@ioc.exchange
0 likes, 0 repeats
@Stormgren ideally you'll be able to characterize the connector launches on your specific board (or use a generic/typical s2p i provide that isn't specific to your s/n) and then do time domain de embedding to a reference plane right at the sfp+ connector footprint (trying to do the connector to optic mating plane would be a lot harder)
(DIR) Post #B6QRJKivyW5HqyN6RM by Stormgren@obsidianmoon.com
0 likes, 0 repeats
@azonenberg That is a heck of a serious board-edge SMA connector! I tend to only use board-edge SMA with short chassis-mounted extension cables for the reason you mentioned, the fatigue tends to kill the solder joint.Back to the decoupling capacitors, I'm also wondering if they are present to account for things like DAC cables that mount zero caps anywhere on them, because those do exist and they're super annoying to work with.
(DIR) Post #B6QRQlE1qEHU4cSQls by azonenberg@ioc.exchange
0 likes, 0 repeats
@Stormgren dacs should, per spec, include dc blocks.On the road but I'll grab the exact line out of sff-8431 in a bit
(DIR) Post #B6QRbxaiComvbwPLTk by Stormgren@obsidianmoon.com
0 likes, 0 repeats
@azonenberg That actually might be somewhat relevant to my work, so thank you!Also, it just occurred to me that for a quick proof of concept while I wait on my osmocom boards is that I could always attach some of these LVDS drivers around here (assuming i can find where i stashed the things) to a cut-up DAC cable to do a quick proof of concept of a thing we want to try.Nutty, but hey, needs must, right? :)
(DIR) Post #B6QSCayfFIKuBs52bg by azonenberg@ioc.exchange
0 likes, 0 repeats
@Stormgren (not saying ones that violate spec don't exist, of course)And it's easier to get a good launch on it with the small center pin than something like Samtec SMA-J-P-H-ST-EM1 with a big chungus that mandates a huge ground plane void on basically any >2 layer stackup. The Rosenberger is a nice coplanar launch and the through board bolts provide mechanical support and allow for some variation in PCB thickness.The amphenol part on the current breakout (don't have the p/n handy since I've stopped using it, the others i use often enough to have memorized) has a tiny lip that doesn't grasp 1.6mm boards well so you put a lot of shear load on the solder when torquing. And unlike the samtec it isn't snug enough to friction fit during assembly which de facto mandates a jig to hold it during soldering
(DIR) Post #B6QSVGiOnxOAqlNP1c by Stormgren@obsidianmoon.com
0 likes, 0 repeats
@azonenberg Oh, I have my copy of 8431 right here, and yes, you are correct, they should have DC blocks on both TX and RX. (Per section 3.4)That being said I have done teardowns on some suspect DACs and the traces weren't even jumpered over where there should be a capacitor, the cap wasn't even there! Wouldn't surprise me if there were some optical units out there that did the same thing for some reason. Initial symptoms didn't present on that issue until someone jumpered the NICs between boxes in adjacent rack cabinets and the NIC crashed in one or both of the boxes.Connection in the same rack didn't trigger the issue in shorter cables but doing a cabinet-cabinet DAC run would. My guess is potential differences based on different power circuits or something, but the cabinets were at least bonded together, so IDK. The solution was to NOT violate our SOP of never doing cabinet-cabinet DAC connections and do something optically, and then rip-and-replace the crap cables.
(DIR) Post #B6QSeO4wnQ8eIJGt72 by Stormgren@obsidianmoon.com
0 likes, 0 repeats
@azonenberg I feel like the Samtec connector's mounting posts are almost structural, like you could hold the board up by the thing if you did it as through-hole somehow. 😁
(DIR) Post #B6QSqPunnJHdSsGJs0 by azonenberg@ioc.exchange
0 likes, 0 repeats
@Stormgren the thing could hold up a small attenuator board or something just from friction without soldering it's super snug.It also *really* wants a good ground on both sides of the board. More of a coaxial launch than a coplanar, if the bottom side legs are unsoldered or attached to floating copper your return loss goes down the toilet. Ask me how i know lol.
(DIR) Post #B6QSz0mOcKG679QdGa by azonenberg@ioc.exchange
0 likes, 0 repeats
@Stormgren (side note, i love how mastodon is the kind of place where i can casually name drop a SFF standard in the middle of a conversation and it's just totally normal and people know what I'm talking about)
(DIR) Post #B6QTBoa7HSDW2zn2MS by azonenberg@ioc.exchange
0 likes, 0 repeats
@Stormgren (this is not an isolated incident I've had separate conversations about 8431 in particular at least twice in the past 12 hours)
(DIR) Post #B6QWhC9C0t4XyUqqbw by Stormgren@obsidianmoon.com
0 likes, 0 repeats
@azonenberg It's a professional hazard. I won't claim to have them memorized, but I keep a few around so I can have arguments with product developers when I have to convince them they're wrong about something (and they're often wrong).That being said, I was just saying something similar before you posted. I never know what sort of crazy discussion I'm going to get into here when I mention something esoteric.
(DIR) Post #B6QX02I6smD6uO0SfY by azonenberg@ioc.exchange
0 likes, 0 repeats
@Stormgren i have definitely quoted 802.3 at people here before