DLTER: A Deep Learning Assisted 3D Environmental Reconstruction Framework in Terahertz ISAC Systems

Terahertz (THz) ISAC which harnesses the key benefits of THz-band in simultaneous ultra-fast transmission and ultra-accurate sensing is one of the crucial enablers for 6G systems. Precise environmental reconstruction is the prerequisite for giving full play to ISAC, but is challenging due to the complex physical environment and certain characteristics of the THz band. In this paper, we explore the potential of deep learning (DL) for environmental reconstruction in THz ISAC, which is one of the pioneer efforts in this direction. Specifically, we propose a novel framework which achieves precise environmental reconstruction by designing two crucial components: THz-based environmental sensing and DL-based environmental reconstruction. To cope with the sparse and irregular environmental data obtained from the first component, we employ a sophisticated DL model in the second component which performs environmental object detection alongside object generation to enhance the accuracy of environmental reconstruction. Experimental results demonstrate the efficacy of the proposed framework in sensing complex environments.