Radio Analytics for Indoor Monitoring and Human Recognition
Liu, K. J. Ray
MetadataShow full item record
In the era of Internet of Things (IoT), researchers have been developing new technologies and intelligent systems to answer the question of who, what, when, where, and how of things happening in the environment. Among the various techniques that measure what is happening in the surroundings, wireless sensing stands out because of its ubiquity and prevalence. On one hand, different indoor activities bring distinctive perturbations to wireless radio propagation. On the other hand, thanks to the nature of multipath, indoor environmental information is recorded and embedded in the wireless channel state information (CSI). Hence, by deploying wireless transceivers to sense the radio propagation environment and analyzing the CSI, one can extend human senses and enrich her/his insight into surrounding environments and activities. By leveraging the natural multipath propagation of electromagnetic (EM) waves, radio analytics is proposed as a promising technology that deciphers radio propagation characteristics and reveals rich environmental information surrounding us. As one approach of radio analytics, time-reversal (TR) technique exploits the information of large degrees-of-freedom delivered by CSI and provides a high-resolution spatial-temporal resonance, by treating each multipath component in a wireless channel as a distributed virtual antenna. The TR spatial-temporal resonance is indeed a resonance of EM field in response to the propagation environment, and it changes whenever the propagation environment changes. Inspired by the principle of TR and motivated by the development of IoT, in this dissertation, we propose several radio analytic systems that leverage multipath information to realize IoT applications of recognizing different events and identifying people in an indoor environment. In the first part, we design three indoor monitoring systems that analyze different event-determined features extracted from either a single CSI sample or a CSI time series. The first proposed indoor monitoring system distinguishes between different indoor events by matching the instantaneous CSI to a multipath profile calibrated in a training database whose similarity is quantified by the time-reversal resonance strength (TRRS). Later on, we derive the statistics of TRRS, and we propose a new TR based indoor monitoring system that differentiates between different indoor events based on the statistical behavior of TRRS. Unlike the previous two indoor monitoring systems which treats each CSI as an independent feature, we propose the third indoor monitoring system by exploiting the temporal information embedded in the CSI time series as an additional feature to comprehensively understand indoor events. Results of extensive experiments demonstrate the proposed systems as promising solutions to future indoor monitoring IoT applications. In the second part of this dissertation, we propose the concept of human radio biometrics and design a through-the-wall human identification system that is implemented on commercial WiFi devices. As a human present in an indoor environment, the radio waves propagate around will interact with the human body through reflection and scattering. We define human radio biometrics as the attenuation and alteration of wireless signals brought by human. We achieve an accurate through-the-wall human recognition by utilizing the fact that the radio biometrics are uniquely determined by the biological characteristics of each human. Through extensive experiments, we validate the existence of radio biometrics and evaluate the accuracy of the proposed human identification system. Unlike conventional approaches for biometric recognition, the proposed radio biometrics system can identify human through a wall and supports commercial WiFi infrastructure, thus illustrating its potential for human recognition IoT applications.