Physical Layer Authentication Without Adversary Training Data in Resource-Constrained Underwater Acoustic Networks

Underwater acoustic networks (UANs) have provided significant support for marine applications. Due to the complex marine environment and broadcast channels, however, UANs encounter huge security threats, which are in dire need of reliable authentication mechanisms. Physical layer authentication (PLA) schemes exploit natural properties of channels, which are incredibly difficult to imitate, resulting in a strong authentication capability. However, capturing a specific channel statistical model from adversaries is quite challenging due to the unknown attacker positions, unpredictable opponent behaviors, and the harsh marine environment. Besides, the limited bandwidth and energy supply of UANs require a cost-effective authentication scheme. In this article, a PLA strategy for UANs is proposed in the absence of prior knowledge of illegitimate nodes, which tries to improve the nodes authentication accuracy with reduced network energy consumption. Two database-correlative features are exploited to jointly characterize the channel impulse response (CIR) patterns over time, which possess strong resilience to the complicated time-varying underwater acoustic channels. Inspired by the spatial dependency property of CIRs in UANs, a training-set construction method without adversary data is proposed to generate the training set for nodes authentication. Using the support vector machine (SVM) algorithm with a small training set, the proposed PLA scheme enables each node to distinct spoofing attackers in a distributed manner. Extensive simulations and analysis based on sea trial CIR data have shown that the proposed PLA scheme has achieved high authentication accuracy with low overhead and reduced network energy consumption.