This is a fascinating subject and one that I'm thinking about at the moment. There are two types of problems - sending serial data between picaxes, and sending programming serial data. This latter is a bit more complex, mainly because the baud rate is a bit higher.
The simplest answer is raw serial data. Manuka has cracked this one with the deceptively simple technique of sending lots of 'U' 01010101 characters before the actual data. Then you can get away with a few characters that are biased with more 1s than 0s or vice versa.
This may well work with induction too. You have a coil and you have a second coil. The volts on the second coil are not referenced to ground - they are an AC signal. But if you bias that AC signal to 50% with a series of 01010101, then you can detect positive and negative transitions after that point. You probably would need some amplification and this might need some op-amps. And I suspect a lot of experimentation on a breadboard with a CRO.
Then there is the more generic "transmit anything" circuit. ie, you put zero on the input and zero comes out the output. This is a modem and if it is possible to get one working, you could do downloads (ie the break signal) as well as data transfers, and it would have applications for RF, infrared and induction.
How low can the carrier frequency go compared with the baud rate? Sure, you can run infrared at 38khz and send data at 1200hz. But can the carrier rate be as low as the baud rate?
Well, I think that might be possible, which could mean that the download baud rate comes within the range of RF modules. Plus it means that one can transmit at the maximum data rate for that piece of hardware.
I've been thinking about Manchester Coding
http://en.wikipedia.org/wiki/Manchester_code
Where this has direct applications is with ethernet, as ethernet is inductively coupled. There is no DC component, so the first design criteria for any code is the number of 1 and 0 has to be equal. This is rather similar to the induction problem posed above.
I'm pondering two circuits. This first is a purely digital design and would need to run at machine code speeds. So perhaps forget that for the moment.
The second uses a D flip flop and an XOR gate to generate the biphase signal from a clock and a data input. Decoding is a bit more complex, but I think it can be done with a couple of very short one-shot timers, an XOR gate, an OR gate and a 4046 phase lock look chip. Lock the PLL by sending a L for a while, which is a series of 1 to 0 transitions. This locks both the frequency and the phase. Then once the phase is locked, sample at 1/4 of the period, after the transition. The 4046 could be used with a divide by 4 counter to generate the right timing point to sample the signal.
This is a bit of a crazy idea. But I've got the cmos chips on order so I think worth breadboarding.
One data protocol that could be very easily implemented is SPI. Clock and data in. And with a PLL you can reconstruct the clock signal and so have clock and data out. But the intermediate data has the same number of 0 and 1 bits, so that makes the datastream perfect for a wide range of AC transmission systems - wireless, audio, induction, light, infrared etc.
Another nice thing about this protocol is that there is no set baud speed. All you need to do is send an initial L which both locks the frequency and the phase. The 4046 can track over a wide range of frequencies with its VCO.
Maybe it might have picaxe applications. The downside is you do need quite a few discrete logic chips (though they are all 25c ones, so the total circuit might still come in less than a single picaxe chip).
Having said all that, if you just want to send one byte, I'd use serin and serout, and use lots of Us at the beginning, then 'ABC', then send the byte and then send its inverse and use that as the checksum. Then the problem is a lot simpler. Just detect the AC signal in a coil. Reference it to 2.5V and just feed the signal into a 1/4 of a LM324, and you should get a digital signal on the output which can go straight into a picaxe. Clamp the receiver coil with diodes so it can't go >5V or <0V.
Hmm- even simpler experiment (which might zap a chip but worth a shot). Reciever coil with one end to the +ve of a 741 and the other end to the -ve. (Mabye 2 clamp diodes on each of those input pins!!). On the +ve transition you get a High output and on the -ve a Low output. I think a 741 can drive a picaxe directly.