Message Authentication Code over a Wiretap Channel

Message Authentication Code (MAC) is a keyed function $f_K$ such that when Alice, who shares the secret $K$ with Bob, sends $f_K(M)$ to the latter, Bob will be assured of the integrity and authenticity of $M$. Traditionally, it is assumed that the channel is noiseless. However, Maurer showed that in this case an attacker can succeed with probability $2^{-\frac{H(K)}{\ell+1}}$ after authenticating $\ell$ messages. In this paper, we consider the setting where the channel is noisy. Specifically, Alice and Bob are connected by a discrete memoryless channel (DMC) $W_1$ and a noiseless but insecure channel. In addition, an attacker Oscar is connected with Alice through DMC $W_2$ and with Bob through a noiseless channel. In this setting, we study the framework that sends $M$ over the noiseless channel and the traditional MAC $f_K(M)$ over channel $(W_1, W_2)$. We regard the noisy channel as an expensive resource and define the authentication rate $ρ_{auth}$ as the ratio of message length to the number $n$ of channel $W_1$ uses. The security of this framework depends on the channel coding scheme for $f_K(M)$. A natural coding scheme is to use the secrecy capacity achieving code of Csiszár and Körner. Intuitively, this is also the optimal strategy. However, we propose a coding scheme that achieves a higher $ρ_{auth}.$ Our crucial point for this is that in the secrecy capacity setting, Bob needs to recover $f_K(M)$ while in our coding scheme this is not necessary. How to detect the attack without recovering $f_K(M)$ is the main contribution of this work. We achieve this through random coding techniques.