Object Imaging With Radio Signals

Imaging technology,which represents the targets’ information such as shapes and locations in the form of images,has various applications including smart homes and target rescue.Radio Frequency(RF)imaging has attracted lots of attention thanks to its all-weather,non-contact,and privacy-preserving characteristics.However,RF imaging faces challenges such as insufficient imaging resolution and complex imaging scenarios,all of which have prohibited existing technologies from practical applications.To address these issues,this dissertation conducts comprehensive research about RF imaging towards high-resolution,multi-scenario,and low-power consumption near-field object imaging,including 1)enhancing the sensing ability by gradually realizing 2D imaging to3 D imaging;2)improving the imaging results through utilizing low-rank nature as priors to further employing deep learning techniques;3)extending application scenarios by studying line-of-sight imaging and non-line-of-sight imaging.The main contributions of this dissertation are listed as follows:1.A Reconfigurable Intelligent Surface(RIS)assisted WiFi imaging system is proposed.To break the spatial resolution limitation constrained by the number of antennas of WiFi devices,the phase shift characteristic of the RIS is utilized to improve spatial resolution.RIS adjusts the reflected signal phase of different RIS elements to generate sharp directional beams for angular scanning.An optimization-based algorithm is proposed,which generates a transform matrix to compensate for the discrete phase error and utilizes prior information of objects to achieve high-resolution two-dimensional imaging.2.A 3-D RF imaging system under the low-rank constraint is proposed.To resolve the problem that 3-D imaging is difficult to achieve due to the narrow band of the WiFi signals,a 2-D array and wide-band signals are utilized to build a system with enough angular and range resolution.Besides,the imaging problem is formulated as an optimization problem with the low-rank nature of objects as prior information to efficiently improve the imaging performance.Furthermore,a coarse-to-fine scan scheme is proposed to efficiently reduce computational complexity.3.An RF imaging system based on deep learning is proposed.To resolve the problem that the imaging performance degradation caused by the increase of target distance,a data-driven image super-resolution architecture is investigated.Networks are designed for extracting contextual information,which is trained with the simulation results,addressing the distortion and fragmentation issues of experimental imaging results.4.A Non-Line-of-Sight(NLOS)imaging system besed on the specular reflection hypothesis with radio signals is proposed.To resolve the problem of imaging occluded objects,the specular reflection characteristic of the visible reflective surface is exploited to establish the signal propagation model,which simplifies the NLOS imaging problem as imaging the virtual object whthin LOS.Besides,a mirror-symmetric mapping model of the virtual targe and the real target is established,and a mapping parameter extraction method without scene prior is proposed.Furthermore,a wavenumber domain imaging algorithm based on confocal approximation is proposed,and the feasibility of non-lineof-sight imaging with radio signals is demonstrated through measurement experiments.