I'm asking if the LEDs that support 9.6Gbps (what is that? A 1ns or faster LED?) even exist in any substantial way.

Isn't that like, laser speeds? Not overhead LED speeds. And over the air? Not inside a fiber line?

This is basically a 10Gbps standard for fiber lines except over the air. All those issues need to be solved without the guarantees or isolation that fiber optic lines provide.

That's just transmit. Receive has to deal with the same issues, amplifiers and other electronics that move at the requested speeds.

Couldn't you just use a laser and a lens?

I'm not an expert in photonics, at all, but I am vaguely aware of the existence of the photonics field in electrical engineering and physics from time in graduate school for electrical engineering. I think that ir-da is probably limited by at least 3 things 1) the extremely simple modulation ir-da uses of basically just turning the IR LED on and off for 1s and 0s - it could be slightly more complex than that for irda, but wifi uses very complex modulation today that is much, much more bandwidth efficient - 2) IR-DA just uses one frequency of light, which kind of makes it somewhat analogous to narrowband RF vs WI-FI which is many MHz wide per channel. There is a ton of bandwidth available in the infrared light space and I'm not surprised at all that they would be able to pull 10Gbps thru light, after all it is much higher frequencies than RF in the first place. And 3) probably IR-DA, being developed in the 90s, uses LEDs and phototransistors that are just simply slow: the transient response of those devices is probably many orders of magnitude slower than what is available today, if I had to guess.

Oh I'm sure that LEDs that support Gbps exist or are about to because vast improvements have been made in basically every corner of the microelectronics space over the years, including sensors and LEDs, not just in computation. It's actually very remarkable what is possible now with many kinds of microelectronics sensors that wasn't possible e.g. a decade ago. The other thing is that probably each LED can operate at a much lower frequency and still get Gbps total bandwidth if a lot of them are used that emit at slightly different frequencies.

Could one PWM such a light to display high bit color content? By my estimates: to send 9.6Gbps, you need (at least) 9.6 GHz, and that's 160,000,000x faster than 60hz. log2(160 million) = 27.3 bits.

I'm sure that this idea is more of a fun hackaday-style gimmick (i.e. not productionizable for various reasons), but would be fun to see.

I don't think they use OOK aka PAM2. I expect something like baseband OFDM fed to the AM-only-capable LED.

Nicely isolates narrow band noise.

Then usual high order QAM in the subcarriers.

256QAM gives you 8bit per symbol.

So it'd only require 125MHz of modulation bandwidth...

You mean 1.2Ghz of modulation bandwidth to reach 9.6Gbps with 256QAM.

Is that a 400ps rise + 400ps fall time LED? Picoseconds?

Or the use of like 8192QAM or something incredible?

I don't know, I was pointing out for the data rate that seems to be part of some COTS devices, not the data rate that you can negotiate with research lab hardware.

Single quantum well LEDs are well-commercialized at 150 Mbps for MOST150 (that was intended to, running over POF, be cheaper than an ethernet PHY and copper); I'm not sure exactly where SQW LEDs drop off and laser diodes become the only option, but it's substantially faster than traditional LEDs.