SLP-Based Dual-Functional Waveform Design for ISAC Systems: A Deep Learning Approach

Abstract

Integrated Sensing and Communication (ISAC) is emerging as a critical enabling technology for 6G communication systems, with increasing attention on dual-functional waveform design to facilitate various applications. Recently developed symbol-level precoding (SLP) demonstrates significant potential for ISAC waveform design due to its superior capacity to enhance both communication and radar sensing capabilities by simultaneously leveraging temporal and spatial degrees of freedom (DoFs). To address the high complexity challenges associated with SLP designs, this paper proposes two efficient deep learning algorithms to tackle the complex SLP-based ISAC waveform design problem. The first algorithm employs a supervised data-driven training strategy that reformulates the original problem as a solvable linear regression issue, while the second algorithm utilizes a gradient-based procedure to fine-tune the trained network, enhancing communication and radar sensing performance. Additionally, we introduce a lightweight SLP-Inception-Net to mitigate computational complexity at the network structural level, incorporating a phase-based activation function to ensure the satisfaction of equality constraints. Extensive simulation results demonstrate that our proposed deep learning algorithms achieve comparable communication and sensing performance to optimization-based approaches, while delivering a remarkable 1000-fold reduction in computational complexity in terms of average execution time. Furthermore, they outperform classical learning-based schemes at the same level of floating point operations per second (FLOPs).