Deep Learning for SLP-based ISAC Waveform Design

Integrated sensing and communication (ISAC) is a key enabling technology for future 6G communication systems. Recently emerged symbol-level precoding (SLP) is considered to possess substantial potential for ISAC waveform design, owing to its ability to enhance multi-user communication and radar sensing performances by simultaneously leveraging both tempo-ral and spatial design degrees of freedom (DoFs). Considering the high complexity challenges brought by existing model-driven optimization based SLP design approaches, in this paper we pro-pose a lightweight SLP-Inception-Net and efficient data-driven deep learning algorithm to solve the highly complex SLP design problem. In particular, we propose a phase-based activation function method to guarantee the equality constraints and use the soft loss to ensure the inequality constraints. Simulation results verify that our proposed deep learning algorithm achieves comparable communication and radar sensing performance to the prior optimization-based approaches, while providing a noteworthy lOOO-fold reduction in computational complexity in terms of average execution time.