Deep Learning Enabled Precoding in Secure Integrated Sensing and Communication Systems

Abstract

This letter investigates the physical layer security of integrated sensing and communication (ISAC) systems, in which a base station (BS) communicates with users and sense targets simultaneously while a eavesdropper attempts to intercept confidential information. We develop an optimization problem for precoding to minimize the maximum eavesdropping signal-to-interference-plus-noise ratio (SINR) while ensuring quality of service (QoS) requirements of communication and sensing. To find its solution, we propose a learning-based precoding scheme that obtains precoding results from uplink pilots and echoes through a neural network without prior channel information. In particular, a newly developed loss function is designed based upon the first-order optimality conditions to take into account the intricate constraints of the ISAC system. Finally, simulation results indicate that the proposed approach effectively reduce the SINR of eavesdroppers while ensuring a satisfactory QoS for communication and sensing performance.