Discriminative Siamese Learning for Pilot Spoofing Attack Detection in MIMO Uplink Systems

Pilot Spoofing Attacks (PSAs) represent a critical vulnerability in Multiple-Input Multiple-Output (MIMO) networks. An adversary transmits spoofed pilot signals to disrupt the channel estimation process at the base station, compromising both system performance and data confidentiality. Recent studies have explored deep learning-based detection methods to address this critical problem. However, existing approaches primarily employ conventional Deep Neural Networks (DNNs), which face challenges in generalizing under varying channel conditions, require large labeled datasets, and often fail to detect subtle attacks, particularly when the eavesdropper transmits at low power. In this paper, we propose the first application of Siamese Neural Networks (SNNs) for PSA detection in MIMO uplink systems. Our data-driven approach learns a discriminative embedding space that enables effective differentiation between legitimate and spoofed pilot signals without relying on fixed thresholds or prior channel knowledge. The proposed framework also provides unified detection of both synchronous and asynchronous spoofing attacks. Extensive simulations reveal that our SNN-based framework outperforms state-of-the-art detection techniques, achieving 17% higher accuracy and 2.2 times less inference latency for a base station with 300 antennas and 20 frames per signal, even under low-power PSAs.