Fighting Against Active Eavesdropper: Distributed Pilot Spoofing Attack Detection and Secure Coordinated Transmission in Multi-cell Massive MIMO Systems

The massive multi-input multi-output (mMIMO) systems are vulnerable to both pilot contamination and pilot spoofing attack (PSA), which jeopardize the uplink channel estimation and cause information leakage in the downlink transmissions. Motivated by the canonical large-scale fading precoding (LSFP) [1], a secure LSFP is proposed to eliminate the impact of the coexistence of pilot contamination and PSA. The proposed framework enables multi-cell coordinated transmissions leveraging the large-scale fading coefficients, thus is suitable to be implemented in mMIMO systems with limited fronthaul. Specifically, the presence of the eavesdropper is first identified by designing a distributed mixture-of-experts neural network (D-MoENN)-based PSA detector, which combines the detected results of distributed nodes to improve the detection accuracy. Subsequently, an optimal jamming base station (BS) is selected by designing a D-MoENN-based localizer, which estimates the locating cell of the eavesdropper and selects the nearest jamming BS to the eavesdropper. Numerical results show that the proposed D-MoENN-based PSA detector outperforms the existing detectors in the low SNR regime. Moreover, the average secrecy rate achieved by the secure LSFP with the jamming BS selected by the D-MoENN-based localizer is close to the upper bound achieved by the genie-aided selector that perfectly knows the location of the eavesdropper.