随着万物互联的发展，无线网络对传感技术的要求越来越高。集成传感与通信技术（ISAC）作为一种新兴的使能技术， 能够有效提升传感效果，为新兴应用提供可靠的支持。然而，现有的ISAC技术在利用无线信号实现高质量环境感知方面存在限制，这阻碍了其在实际系统中的应用。因此，本文专注于研究面向无线传感的室内环境感知关键技术。 首先，针对被动传感的单天线超宽带无线通信场景，本文提出一种基于多径时延的环境感知方案。该方案基于已有的同步定位和建图（SLAM）概率模型，考虑无线信号传播中反射和散射效应，提出一种两阶段机制求解与发射机关联的镜像点和散射点坐标。进一步，基于所提出的三维（3D）环境感知算法实现室内3D点云感知。实验表明，与多天线方案相比，所提出的方案在移动发射机位置估计误差较小的情况下可以显著提升感知精度，并且在信噪比（SNR）恶劣的环境中具有更优的感知性能。 接着，针对被动传感的单天线窄带无线通信场景，本文提出一种基于频率响应的环境感知方案。为解决不同空间位置观测的频率响应的环境特征提取问题，提出一种感知关联注意力机制。此外，考虑到不同空间位置观测的频率响应在不同排列组合下所表征环境特征的一致性，提出一种随机排列策略。在此基础上，提出一种深度学习模型Radio2Vox，用于实现室内3D环境感知重建。实验表明，所提出的模型对整体环境重建的交并比（IoU）最高可达98.9%，对内部环境细节重建的IoU最高可达93.5%。此外，对比实验显示，该模型在此任务上具有明显的优越性能。 最后，为了应对主动传感中硬件效益与感知精度平衡的挑战，本文提出一种基于虚拟孔径的环境感知方案。该方案利用驱动装置驱动单发单收超声波雷达运动，从而形成更长的孔径，并结合所提出的非线性感知算法实现高精度感知。为克服超声波雷达换能器与麦克风在物理空间分置给系统模型分析带来的挑战，引入了等效虚拟收发一体化的概念。仿真实验评估了算法的可行性和性能。同时，原型实验表明所提出的方案可在米内范围实现厘米级的分辨率性能，且可以感知障碍物的横截面大小。
As the Internet of Things (IoT) evolves, the demand for sensing technology in wireless networks is increasing. Integrated Sensing and Communication (ISAC) technology, as an emerging enabling technology, effectively enhances sensing capabilities and provides reliable support for emerging applications. However, existing ISAC technology faces limitations in achieving high-quality environmental sensing using wireless signals, which hinders its practical application in systems. Therefore, this thesis focuses on the research of key technologies for indoor environmental perception with a focus on wireless sensing. First, for passive sensing in single antenna ultra-wideband wireless communication scenarios, this thesis proposes an environment perception scheme based on multipath delay. This scheme, based on existing probabilistic Simultaneous Localization and Mapping (SLAM) models, considers reflection and scattering effects in wireless signal propagation, and proposes a two-stage mechanism to solve for the coordinates of mirror points and scattering points associated with the transmitter. Furthermore, based on the proposed three-dimensional (3D) environment perception algorithm, indoor 3D point cloud perception is achieved. Experiments show that compared to multi-antenna comparison schemes, the proposed scheme can significantly improve the perception accuracy under small transmitter position estimation errors, and has better perception performance in environments with poor signal-to-noise ratio (SNR). Next, for passive sensing in single-antenna narrowband wireless communication scenarios, this thesis proposes an environmental perception scheme based on frequency response. To address the problem of extracting features from the frequency response observed at different spatial positions, an aware relational attention mechanism is introduced. In addition, considering the consistency of environment features represented by the frequency response under different spatial positions and permutations, a random permutation module is proposed. Based on this, the deep learning model Radio2Vox is proposed to achieve indoor 3D environmental perception and reconstruction. Experiments show that the proposed model can achieve a maximum IoU (Intersection over Union) of 98.9% for overall environmental reconstruction and a maximum IoU of 93.5% for internal environmental detail reconstruction. Furthermore, comparative experiments show that the model has significantly superior performance in this task. Finally, to address the challenge of balancing hardware advantages and perceptual accuracy in active sensing, a virtual aperture-based environmental perception scheme is proposed. This scheme uses a drive mechanism to move a single-transmit, single-receive ultrasonic radar to form a longer aperture and combines it with a proposed non-linear perception algorithm for high-precision sensing. To overcome the challenges posed by the physical separation of ultrasonic radar transducers and microphones for system model analysis, the concept of equivalent virtual transceiver integration is introduced. Simulation experiments evaluate the feasibility and performance of the algorithm. Meanwhile, prototype experiments demonstrate that the proposed scheme can achieve centimetre-level resolution performance within metres and can perceive the cross-sectional size of obstacles.
