DNN-Based Detection of Pilot Spoofing Attacks in Massive MIMO Networks with Phase Noise

Massive multiple-input multiple-output (MIMO) networks are known to be susceptible to pilot spoofing attacks (PSAs), in which an active eavesdropper (ED) sends the same pilot signal as that of the attacked legitimate user equipment (UE) during the uplink channel estimation phase. A PSA causes information leakage to the ED and also weakens the received signal strength at the attacked UE. We assume the practical case of non-ideal local oscillators that introduce phase noise (PN) at the base station, UEs, and ED. We show that in the presence of the ED, the PN increases the rank of the signal covariance matrix by one, which is exploited in the detection of PSA. We propose a deep neural network, called attack detection network (ADNet), to detect the PSA by exploiting the eigenvalues of the received signal sample covariance matrix and the power ratio as input features. Numerical results show that the proposed ADNet is effective in detecting the PSA and reveal that the larger the PN, the higher the detection accuracy.