Joint Design for RIS-Aided ISAC via Deep Unfolding Learning

Abstract

Integrated sensing and communication (ISAC) has become a promising technique to alleviate the spectrum congestion via sharing the same spectrum for communication and sensing. Nevertheless, many ISAC schemes encounter the challenges of high computational complexity. Thanks to the powerful non-linear fitting capabilities and fast inference speed, deep learning is expected to facilitate the online deployment of ISAC. In this paper, we propose a dual-functional waveform design scheme for reconfigurable intelligent surface (RIS) aided ISAC based on deep unfolding learning. Specifically, the weighted sum of multi-user interference energy and waveform discrepancy is minimized via the joint waveform and phase-shift design. We first develop an alternating direction method of multipliers (ADMM) based iterative algorithm to handle the non-convex optimization problem. Then, we develop a deep unfolding neural network (NN), named ADMM-NET, which unfolds the proposed ADMM-based iterative algorithm to a layer-wise architecture and replaces the matrix inversions with low-complexity approximations. In addition, we present a black-box NN for performance comparison. Simulation results verify that the ADMM-NET outperforms the black-box NN in performance, interpretability and training samples. Moreover, the ADMM-NET is superior to the ADMM-based iterative algorithm in both computational complexity and performance, facilitating the online deployment.