Research On Device-free Wireless Sensing

Recent advances have shown that when a person is in a wireless network,his or her behavior has complex influence on wireless signals which could be exploited and utilized to sense the person's location,activity,gesture,identity and moving trajectory.Since it does not require human to equip extra devices,we call it device-free wireless sensing(DFS),which propels the evolution of traditional wireless networks into smart networks.We try to sense the target's location and activity by the analysis of wireless signals influenced by the target.Compared with traditional statistical features which provide only partial information to characterize the received signal strength(RSS),this paper explores the methods to characterize the distribution of the signals as a whole,exploits distribution features to obtain more useful information,so as to improve the performance of device-free wireless localization and activity recognition.We develop a novel coherence histogram to better characterize the distribution of RSS.It captures not only the occurrence probability of RSS measurements,but also the spatial relationship between adjacent RSS measurements.Due to the spatial structural information,coherence histogram is proved to be an efficient and informative representation of the raw RSS measurements.To improve the performance of DFS systems in complex scenarios,such as through-wall or non-line-of-sight scenarios,we try to extract more useful information to characterize the influence of the target from wireless signals.In this paper,we deal with the two-dimensional channel state information(CSI)measurements on multiple channels as a whole from a multidomain point of view.We adopt the distribution of structure blocks in both the time domain and frequency domain to characterize the CSI measurements,so as to extract more informative features involved not only in the time domain and frequency domain,but also in the spatial structural domain.We develop two schemes to preserve the signal features in spatial structural domain.On one hand,we partition the measurement matrices into basic structure blocks,adopt self-organizing map networks to cluster the blocks into a number of categories,so as to make it feasible to characterize the block distributions as a whole using a probability distribution function.On the other hand,we adopt the coherence histogram to represent the distribution of the blocks,which considers not only the occurrence frequency of blocks that belongs to a certain category,but also the spatial relationship between adjacent blocks.Wireless letter recognition is a novel application in the field of DFS,which realizes letter recognition in the air utilizing wireless signals influenced by the hand's movement.However,when the target writes different letters,the range of the hand's movement is very limited.And it is hard to extract distinguishing features,since different letters are very close to each other.To solve this problem,we build a wireless letter recognition system.Specifically,we design a differential method to extract robust CSI measurements from the noisy wireless signals,develop a variance based method to detect the start and end point of the letter writing activity,and utilize the coherence histogram based multi-domain feature extraction method to extract discriminative features for letter recognition.We have done extensive experiments to verify the feasibility of the proposed methods above.It can be observed from the experimental results that our proposed methods achieve outstanding performance.