Simplex and simple?  Can you put them close enough together that the light from the LEDs will dominate ambient?  

If so, I would use two LEDs on the transmit side, two phototransistors/diodes on the receive side.  Run "half-duplex SPI" -- one of the LEDs is clock, the other data (MOSI, I guess).  You can piggyback on the chips' SPI hardware. This could be very easy.

If the background is bright, you're probably going to have to modulate the signal like with normal 38 kHz IR, and then maybe radio is better.  nRF24s if you need speed?  

half-SPI is a great call. Will see if the LEDs can overcome ambient. It's indoors at least. I was thinking about making it differential (two LEDs, one signal line with the UART plan, or 4 LEDs with half-SPI) so that only when one is LED from a pair is detected but the other isn't we have a deliberate signal, as opposed to when the ambient lights get bright and both photodetectors register a 1. Then we get into AGC and I realise we may want to go back to prebuilt IrDa or 3XkHz modules as things are getting hard again!

Maybe use infrared modules? It's exactly your idea of splitting an opto-coupler, but with a bunch of the mechanical and electrical engineering already done. Here's one datasheet, but there seem to be quite a few around: http://www.vishay.com/docs/82633/tfbs4711.pdf It claims to support 115k baud.

Splendid. Saves rolling our own for the first run. Price per unit could perhaps be bested by discretes over short range. I like the configurable current for the LEDs. Thank you.

Have a look at: https://www.pjon.org/AnalogSampling.php

This is fantastic - github repo too! Thanks 

Nice! This is very much what I could use for my data transfers from the bee hives as well!

Thanks for the link!

How about piggy backing on something like QI charging, but modulate it like a NFC signal?  Before I decided to go with an ESP-12F this was a path I was preparing to venture down.  Solves getting power to the device across that air gap as well.  A single wire, air core split inductor should give you a data rate somewhere from 0.5 to 2mbit range from my back of the envelope calculations.

Thanks Daren, I am attracted to the idea. Have you seen anything similar? Is it anything like data modulated over mains AC? Or NFC (Manchester encoding 10%, low pass filter)? 

Not sure what the theory of operation would be (send power over Qi, "power Rx" device [which is the the "data Tx" device] somehow shorts or uses caps to pull the coil voltage, "data Rx" uses increased/reduced current sense across a resistor to pull signal out of the Qi Tx coil and then puts through another LPF to clean up, then opamp before reading with the uC?) , so any references would be gratefully received. 

Well the Qi standard itself includes data transfer capability, but the data rate for that is only about 2kbit.  Mainly because... to operate at greater efficiency levels, the Qi system is designed to operate at resonance and then the data transmission actually occurs by de-tuning the coil from that resonance... kinda like FM radio, but you don't have a set center frequency!  (as far as I understand it anyway)

The target for that resonance was also established to be a pretty low... 140kHz.

However, most DC-DC buck/boost converters, which essentially operate on a very similar principal, often resonate at 2MHz or more.  The inductor required then gets smaller and smaller as well.

So you would be in a bit of uncharted territory I think, but it could work...  How much you need to deviate from Qi would depend on the data rate you need.

I like the idea of an air core transformer for a high frequency switching regulator! I only need single digit mA on the data Tx side. 

Much to look at on the Qi standard to find out what techniques are used to read "detuned" resonant magnetic fields. 

Thanks