Network error correction coding for time-varying adversarial errors in a unicast network

We consider a unicast network with an adversary who is able to attack a proportion of edges, which is no more than p for 0 < p < 1. Based on the knowledge of the adversary, three cases of attacks are considered: 1) the adversary knows nothing about the source message; 2) the adversary knows the source message but does not know the transmitted codeword; 3) the adversary knows the transmitted codeword. Two classes of code schemes, deterministic code and stochastic code, are designed to assure reliable transmission against the adversarial attack. It is concluded that stochastic code does not provide any more benefit than the deterministic code against the attacking cases 1 and 3. However, when the attacking case 2 is considered, stochastic code would achieve a higher capacity than the deterministic code.