ISAC-NN: A Deep Neural Network for Efficient Downlink Beamforming in Integrated Sensing and Communication Systems

The rapid expansion of wireless devices has led to severe spectrum congestion, driving the need for innovative solutions that integrate sensing and communication functionalities. In this paper, we propose integrated sensing and communication neural network (ISAC-NN), a lightweight deep neural network designed for downlink beamforming in ISAC systems. By modeling the received signals at communication users alongside the echoes from sensing targets, we establish a relationship between the beamforming vector and the signal-to-noise ratio (SNR). Uniquely, our approach incorporates the SNR contributions from both communication and sensing in the network’s loss function and employs an offline learning strategy to train a neural network model that infers the optimal base station beamforming vector. Extensive numerical simulations across various antenna array configurations demonstrate that ISAC-NN not only facilitates effective beam design but also achieves a balanced trade-off between communication and sensing performance. Moreover, the proposed method outperforms conventional techniques, particularly in large-scale array scenarios, underscoring its potential for deployment in future 6-th generation mobile networks (6 G) and Internet of Things (IoT) wireless networks.