Smart User Sensing Based On Passive Wireless Backscattering

In recent years,with the empowerment of emerging technologies such as artifi-cial intelligence,edge computing,5G,and the large-scale deployment of mobile smart devices,the development momentum of the Internet of Things(IoT)has been continu-ously enriched.The development speed of IoT has been accelerating,and the potential market space is vast.Still,sensors under the IoT are numerous and widely distributed.Traditional battery-powered energy maintenance costs are high,making the IoT limited by energy issues.In recent years,the theoretical research of passive reflection commu-nication technology has gradually deepened.It has realized the use of energy in the environment to carry out data transmission and information interaction through the op-erating mode of signal reflection.Passive networks empower the IoT,which will realize the transformation of the IoT from active to passive and derive more efficient and conve-nient application scenarios.After being widely deployed,the passive networks provide opportunities for user sensing while completing the communication and transmission of their tasks.User sensing is the link that connects the physical world and the information world.It promotes the interaction between people and things.As a core capability,it supports various IoT industry application cases,such as smart homes,smart offices,and medical care.In recent years,people have proposed sensing systems based on various means,but they may have privacy issues,require additional equipment to be worn,or may be inter-fered by other users and wireless devices.To solve these problems,we use passive re-flected signal to explore low-power,anti-interference,non-intrusive and high-precision multi-person intelligent sensing.In this article,we propose BackGuard,a multi-person sensing system in IoT scenarios,which can realize accurate and non-intrusive activ-ity recognition and user identification.BackGuard collects backscattering signals and analyzes the impact of user behavior on signal transmission.It helps extract the physio-logical and behavioral characteristics embedded in the user’s daily activities.To extract meaningful activity segments from the collected signals,we design a signal processing method and an adaptive segmentation algorithm.Then we propose a parallel attention-based deep neural model,which integrates two cross-domain labels,activity and iden-tity,to improve training efficiency while ensuring accuracy.We built the BackGuard prototype system and created a large-scale user data set for the daily behavior.Using the data set,two-month experiments were carried out on the system BackGuard.Differ-ent environmental parameters were set in the experiment,such as the number of people,distance,and time.Experiments show that the system can achieve 93.4%activity recog-nition accuracy and 91.5%user identification accuracy.The accuracy of the system can still be guaranteed in the multi-person scenario.