Generative AI for Sparse Antenna Array Design in ISAC Systems

The fusion of sensing capabilities into communication systems stands as a pivotal advancement in wireless networking, giving rise to the paradigm of Integrated Sensing and Communication (ISAC) systems. In addition to the need for high performance in both sensing and communication functions, ISAC systems require energy efficiency and cost effectiveness. Antenna selection emerges as a strategic means to reduce power cost. The current work proposes a novel optimal antenna selection framework for ISAC systems, that is predicated on the deep generative modeling paradigm of Generative Flow Networks (GFlowNets). Within this framework the antenna selection process is modeled as a deterministic Markov Decision Process (MDP) where sensor subsets arise as terminal states. Given that the objective comprises a convex combination of two terms, one for sensing and the other for communication, a multiobjective formulation is employed. This formulation conditions GFlowNet training on the vector of coefficients for both terms. The proposed approach not only optimizes the collective performance of the systems but also demonstrates generalization to scenarios with coefficients that are not used during training.