Well thats interesting,
assuming you have the right 'interfacing hardware' you could do as Mark Lord suggested and decode the incoming bitstream in software inside the kernel.

Of course, the messages sent down this wire, won;'t just be 'mute head unit', 'unmute headunit', and there may be quite a lot of traffic down this wire to the headunit, so any kernel decoding would have to be (a) real time (or as near as can be achieved) and (b) able to decode the PWM bit stream into a binary string of bits [This would make it timing sensitive].

Once that was done Mark could just take the 'bit stream' and turn it directly into a 'fake button code' for each message (this would be a different fake button code for each unique bit stream message.
You could then use this fake code in your IR Maps without needing to know the exact message format - simply treat the relevant ones [like mute headunit, unmute ...] as button codes and act accordingly.

However I am not sure if Mark (or anyone else on the BBS) has a car with Onstar so I am not sure how we could get some data to play with to analyse it further.
We also don't know the voltage of the signals down this wire, I'd assume 5 volts TTL but thats only a guess.

One thing I can say, if you want to progress this part further you need to find someone who has access to a Onstar equipped vehicle. [and probably the ability to hack the kernel in the empeg].