Hacker News Reader

Comment by fru654 | Hacker News Reader

I wonder if something like this is possible with HDMI? Separate 10G SFP+ for each color channel, one more for i2c, create a similar style breakout PCB, maybe add an MPO or CWDM mux… Could be a fun project. Optical HDMI cables are expensive and most of the time come with a preexisting cable which is hard to route (in conduits) due to HDMI connector size.

View on HN

Such products are already commercially available [0][1]!

DIYing it is probably too painful to be doable. You won't be able to source any kind of protocol translation chip, so you'll have to send it essentially raw into quad SFP+ transceivers. Running 4+ fibers instead of the required 2 (or even 1) is very expensive, and any kind of WDM immediately blows up your budget. Unless you're getting the stuff for free from a DC renovation or something, it's just not worth it.

On top of that you also have to deal with designing board for extremely fast signals, which is pretty much impossible to debug without spending "very nice car" amounts of money on tooling. People have done it before, but I definitely don't envy them.

[0]: https://www.startech.com/en-us/audio-video-products/st121hd2...

[1]: https://www.blackmagicdesign.com/products/miniconverters/tec...

If you need 4x channels, it sounds like a job for QSFP? HDMI is already differential signalling, so you don't need to do that, but you might still need level shifting.

Probably a box on the source end to manage DDC and strip HDCP.

> You won't be able to source any kind of protocol translation chip

I think many of those chips are simple off-the-shelf parts. Probably you would need special licenses only to decode HDCP.

If you have an FPGA, you could even create valid Ethernet frames and send the data / video stream over any standard switch / media converter as long as you have enough bandwidth and no packet loss. (10G would be enough for FullHD and 25G for 4K if you make it a bit smarter and can strip the blanking interval.)

There's even cheaper versions of this now, "fiber" HDMI cables with the electronics in the HDMI plugs themselves, no additional power required. They go up to 100m length. I do wonder how these work, since I've never seen a good teardown of one.

> You won't be able to source any kind of protocol translation chip

This is called an FPGA.

My plan, if I ever need long haul(>3meters) video or audio links, is to get the signal into ethernet(or even better ip) and use common network equipment to transport it.

The theory being ethernet is such a well developed, easy to source common jelly-bean part that this would trump any gains that specialized transports might otherwise have.

But this is probably just my inner network engineer being disdainful over unfamiliar transport layers.

Nah, this is totally the reasonable way to do it, iff you can tolerate the compression loss or whatever. Because 4k60 is like 12Gbps uncompressed, and even more after you cram ethernet headers onto everything. So most such devices include some compression, and the really expensive ones let you configure how much.

Failing that, you're probably doing SDI over your own lambda.

That is happening in the pro world, check out e.g. SMPTE ST 2110.

It's much cheaper to just buy an optical HDMI cable if you need a long point to point run, it's like 50 bucks for 100 ft. The cool stuff you can do with HDMI over IP lies in switching the signal to different endpoints on demand and things like multicast to multiple receivers, both of which are things you can do with off the shelf HDMI over IP gear.

I have wondered about the same (and/or DisplayPort) but with QSFP optics to simplify dealing with the four channels of data.

"Classic" DVI-derived HDMI would probably be trickier because of variable clock speeds and additional data but modern HDMI 2.1 is pretty similar to DisplayPort in that it uses four lanes at fixed rates and sends data as packets over those.

I would love to be able to use standard widely available fiber patch cables for long distance video runs rather than needing proprietary cables only offered in fixed lengths and equipped with enormous connectors that are not friendly to conduit.

Also these days data rates are getting high enough that even normal lengths are problematic, DisplayPort just recently announced that 3 meter cables will need active components for the full 80 gigabit per second mode, which means that a computer on the floor connecting to a monitor on a standing desk will not be guaranteed to work with passive cables. HDMI also recently announced version 2.2 with a bump from 48 to 96 gigabits per second so they'll presumably be in the same boat.

Not HDMI, but SDI over fiber is basically this. It can be muxed and is used in the broadcast industry for long haul camera feeds.

SDI over fibre with a cheap converter if you need to push multi hindered metres. Then people moves to 222-6 which packetised the SDi over IP, and now 2110 which breaks out the SDI to its components.

For most long haul links people still compress, good old h264 or 265 with latencies in the 200-500ms range (plus network propagation), or J2k/JXS and NDI which are more like 50-150ms. Ultimately 200mbit of h265 is far cheaper to transmit than 10ish gbit of 2110, and in many cases the extra 500ms doesn’t matter.

Reply on news.ycombinator.com