coding theory: CRC hamming distance, non-trivial real world situation

The problem: I am jointly developing a low-speed automation bus for low-cost hardware and I need some error verification for your messages. Using a CRC for that seems to make sense, but which one?

I found an article by Koopman + Chakravarty but it doesn't say how they calculated Hamming's distances from the polynomials used.

An additional complication is that our bus design offers three possible values ​​that you could use to calculate a CRC:

  1. the real binary message
  2. the state of the $ n $ bus cables in each time slot
  3. as above, but XOR to the previous slot: it cannot be zero

Algorithm to encode a message: given $ n $ wires there $ s = 2 n-1 possible states (one is excluded because it is identical to the previous one). Therefore, treat the message as a bit stream, divide it into 11-bit or 14-bit boxes, treat each box as an unsigned number, convert it to base $ s $, add one to each digit, XOR the binary value of each digit in the state of the wires, wait a couple of µsec, repeat. The receiver simply reverses this process.

I would like to discover how to achieve a maximum distance from Hamming to $ le 100 $Bit messages, with the least amount of overhead (that is, a CRC-8 if possible), and I freely admit that I have more questions than answers at this time.

For example, the state of the XOR & # 39; d bus has the interesting property that, by definition, there are no single-bit errors and there are no long series of zeros: does that affect the importance of CRC? how? Koopman + C. provides maximum Hamming distances for several "good" polynomials and block lengths, but a 1-bit error in the hardware state changes from 1 to ~ 10 bits in the resulting message: this obviously affects the end behavior At the end of the CRC, but is it possible to select a polynomial so that it cannot result in a Hamming distance of 2?

NB: 11-bit and 14-bit frames are used because they are at most 16 bits long (important for a fast 8-bit CPU implementation) and have a low overhead, i.e. $ 7 ^ 5 $ is reasonably close to $ 2 14 If the bus has three wires. Similarly, 11-bit blocks are used for 2 and 4 wire buses.

A brute force approach would be to encode random messages, protect them with a selected CRC pair, bombard them with 1 … 5 random bit errors, do this on a large machine with several CPUs for a week and see where I find false positive …

If anyone has a more efficient idea, I would appreciate it.