上海科技大学知识管理系统

The integrated sensing and communication (ISAC) technology, through its efficient collection of sensory data, can empower higher-layer artificial intelligence (AI) tasks, such as motion recognition, environmental monitoring, etc. In this paper, we consider that the ISAC transmitter trains its learning model using the noisy sensory data sensed from the environments, while communicating with the communication users. On one hand, during sensory data collection at the ISAC transmitter, different sensory data are subject to varying degrees of sensing errors. On the other hand, different sensory data contribute differently to the learning tasks. Therefore, when evaluating sensing performance, it is necessary to consider both the importance of sensory data and the impact of sensory data errors on the learning tasks. In this paper, we propose a data-aware beamforming scheme to optimize the performance tradeoff between communication and sensing, where new data-aware sensing metric is adopted to guarantee the accuracy of model learning. Experiments prove that through data-aware resource allocation, the proposed scheme can achieve better performance tradeoff between communication and learning task compared to baselines.