Finally, close to 5am after a long day of hacking, I was able to make the first actual voice calls through OpenBSC. This has been _much_ harder than anticipated, with bugs spread all over various places of the code (my code!) as well as some conceptual shortcomings in my understanding of how the exact interaction between GSM 12.21, 08.58 and 04.08 happens.
Initially I was working with the ip.access BTS, but was stuck and didn't see any progress so I decided to try the BS-11, since that one is at least according to spec as A-bis over E1 is publicly documented. Some six hours of problems in my E1 sub-channel multiplexer and TRAU decode/recode later, I actually had a working voice call. To my big surprise, the delay between two phones at the same BTS is incredibly long. Feels like VoIP between two continents.
I then tried whether the fixes have changed anything for the ip.access situation. And yes, they did! It was working straight away, and the delay is non-existent. Isn't it ironic that the traditional E1 system is much worse than the fancy new IP based system? I really need to hunt down where the delay is introduced in the E1 TRAU frame handling, this is not really acceptable right now.
Some further experimentation discovered that my Motorola EZX phones somehow refuse to work, while all the Nokia ones do. Well, some more bugs to fix down the road. At least conceptually everything is working now.