Deep Learning for Detection and Identification of Asynchronous Pilot Spoofing Attacks in Massive MIMO Networks

Massive multiple-input multiple-output (MIMO) networks are highly vulnerable to an active eavesdropping attack called pilot spoofing attack. The pilot spoofing attack causes information leakage to the active eavesdropper (ED) and also weakens the strength of the signal received by the attacked legitimate user equipment (UE) during the downlink transmission. In this paper, a deep neural network, called identification network (IDNet), is proposed to detect asynchronous pilot spoofing attacks and identify the attacked UE. We show that an asynchronous pilot spoofing attack leads to increasing the signal subspace dimension by one unlike the synchronous one. This property is then exploited to improve the attack detection/identification accuracy. In the proposed IDNet, the input features are the eigenvalues of the sample covariance matrix of the received signal at the base station (BS) as well as the ratio between the power of the received signal at the BS projected onto the pilot signals and its expected value. Numerical results show the effectiveness of IDNet in identifying the attacked UE and reveal that the larger the timing and/or frequency mismatches of the ED, the higher the identification accuracy confirming that asynchronous pilot spoofing attacks can be identified more accurately than synchronous pilot spoofing attacks.