Detection of channel degradation attack by Intermediary Node in Linear Networks

We consider the problem of two sources wanting to share information through a potentially untrustworthy intermediary node. We assume that the two sources transmit random symbols simultaneously and that the intermediary node relays the information in the amplify-and-forward manner. We show that under a certain sufficient condition on the channel, it is possible to asymptotically detect whether or not the intermediary node is degrading the channel by sending out manipulated symbols. This can be done solely by each source examining its received distribution conditioned on what it transmitted; thus allowing for a minimally invasive approach to determining if the intermediary node is acting maliciously. More specifically, we model the potential malicious action of the intermediate node by an “attack” channel. An estimate of the attack channel is obtained from the received conditional distribution empirically observed by a source node. We show that the estimated attack channel converges in probability to the true attack channel if the intermediate node is not acting maliciously. Otherwise there is a separation between the estimated and true attacking channels with high probability. This result provides us a clear-cut criterion to determine whether the intermediate node is malicious or not.