Magnus,

Surviving hostile communication links is the whole challenge of designing radio modem links. Its two orders of magnitude more difficult than cabled interfaces.

Fifteen years ago I did this sort of work as a specialty. I'd go into a company and work with a Radio Engineer in their transmission lab. With all their expensive lab equipment, it would take from 9 to 18 months to design, program and certify the system.

The bad news is that I can't think of any EASY advice to offer. The methodology is very dependent upon what you have in your lab, particularly your radio link. The whole process is about punching a signal through chaotic conditions and its quite involved dealing with all the firmware components for error detection/correction, packet protocols, reconstruction strategies etc. Its not trivial.

Maybe someone can suggest an easier technique.

HISTORY:
- - - - -
What we did would be considered basic in today's technology. Everything was purely microprocessor bit-banging; everything was read into the micro from a zero-crossing detection digital toggle signal and everything was sent as a timed square wave into a sineusoidal shaping circuit. We used a quibit encoding system and mountains of message reconstruction firmware. We operated at 4800 baud which at the time was remarkable. The code listings were about 2.5 inches thick in 8085 assembly language.

To make matters worse, the contract required us to get a message through a "10% off transmitter duty cycle to simulate worst-case noise in the field."

The chief engineer advising the client wanted to use a Motorola design and he "fixed the contract against us" after his initial trials amongst three companies. Motorola used a blurb-packet technique that sent and multiply resent a bunch of 4 ms packets that built up into one. We sent a 32 ms minimum packet and used crcc and error correction/detection to protocol the resend handshake.

When the trial came around, the chief engineer advising the client INTERPRETED the "10% off transmitter" to mean for the sake of the trials, "off for 1ms, on for 9ms" thereby ensuring that no complete packet of our ever got through unmolested. They gave us 4 months to respond.

We added inline resync cues and message reconstruction by partials, and a branch and bound cumulative CRCC checker that tried all possible rebuild permutations in the time the previous CRCC ran a single pass. Our system delivered system information quicker and more reliably in comparison to Motorola. The third contender simply dropped out when they defined the 10% trial.

It was a beautiful system. We won the contract and I still see these system in use in California. Motorola probably ended up with the market after that company was sold.

I guess the lesson is, "IT AIN'T EASY"

-Jay C. Box