On Detecting Pilot Attack in Massive MIMO: An Information-based Clustering Approach

In time-division duplex massive MIMO systems, the downlink channel state information (CSI) can be obtained at the base station (BS) from uplink training owing to channel reciprocity. Pilot contamination attacks may occur during the training phase, where the adversary user sends the same pilot sequence as that of legitimate users so as to deceive channel estimation. In this paper, we study the pilot attack detection problem in the time-varying channel environment (e.g., drone networks). Inspired by rate-distortion theory, we propose a theoretical framework to detect such pilot attacks, by counting the number of distinct underlying distributions in the received pilot signal. Information-based clustering methods are designed to improve the detection accuracy by exploiting block-sparsity in data structures. The simulation results show that the proposed approach works successfully in the time-varying environment.