4 TsinghuaScienceandTechnology,February 2015, 20(1): 1-6
consider multipath harmful, since RSS is unable to resolvemultipathpropagationandsuffersunpredictable fluctuation in dense multipath propagation. In contrast, CSI manages to resolve multipath effect at subcarrier level. Though coarse-grained, CSI offers opportunities to harness multipath in wireless sensing applications. 3.1 Sensingtheenvironment In multipath environments, propagation paths can be broadly classified into Line-Of-Sight (LOS) and NonLine-Of-Sight (NLOS) paths, where NLOS paths often pose major challenges for wireless communication and mobile computing applications. Severe NLOS propagation may deteriorate communication quality and degrade theoretical signal propagation models. A prerequisite to avoid the impact of NLOS propagation is to identify the availability of the LOS path. Since CSI depicts multipath at the granularity of subcarriers, researchers have explored CSI for LOS identification[14, 15]. Zhou et al.[14] extracted statistical features from CSI amplitudes in both the time and frequency domains, and leveraged receiver mobility to distinguish LOS and NLOS paths based on their difference in spatial stability. Wu et al.[15] utilized CSI phases of multiple antennas for realtime LOS identification for both static and mobile scenarios[15]. Phase information offers an orthogonal dimension to traditional amplitude-based features, and has been successfully adopted in a range of applications, e.g., millimeter-level localization[16]. Another concrete environment characteristic is the shape and the size of rooms and corridors, which make up part of the floor plan. Floor plan is often assumed to be offered by service providers and researchers have shown increasing interest to draw floor plans by combining wireless and inertial sensing. Some works also demonstrated the feasibility of using wireless sensing alone to recover part of the floor plan information. For instance, Wang et al.[17] distinguished straight pathways, right-angle, and arc corners by analyzing the difference in the trend of CSI changing rates when the WiFi device moves. With channel measurements on multiple receiving antennas, the authors in Ref. [18] developed a space scanning scheme by calculating the angle-of-arrivals of multiple propagation paths and inferring the locations of the reflecting walls. Despite its bulky size, the working prototypeholdspromiseforscanningthephysicalspace
wirelessly and contactlessly. 3.2 Sensinghumans Humans, as part of the environments wireless signals propagate within, are of utmost interest in wireless sensing. Inpassivehumandetection,CSIcandetecttiny human-induced variations from both LOS and NLOS paths, thereby enhancing detection sensitivity and expanding sensing coverage. Zhou et al.[4] utilized CSI as finer-grained fingerprints to achieve omnidirectional passivehumandetectiononasingletransmitter-receiver link, where the user approaching the receiver from all directions can be detected. With fusion of multiple links, CSI also facilitates fine-grained passive human localization[19].Xi et al.[5] extendedhuman detection to multi-user scenarios by correlating the variation of CSI to the number of humans nearby for device-free crowd counting. Pioneer research has marched beyond detecting simply the presence of humans. On the one hand, CSI-basedwirelesssensingshiftsfromlocatingusersin thephysicalcoordinatestoofferingmorecontext-aware information. Some work demonstrated the feasibility of general-purposed daily activity recognition by using CSI as fingerprints for the hybrid of locations and activity patterns[7]. Others targeted at more concrete scenarios, e.g., fall detection[20], adopting similar principles with scenario-tailored optimization. On the other hand, ambitious CSI-based sensing applications strive to detect micro body-part motions at increasingly finer granularity. Some reported over 90% accuracy of distinguishing multiple whole-body[6] and bodypart gestures[9], while others claimed accurate breath detection[10] or even lips reading[8]. Nevertheless, researchershavereachednoconcursusontowhatextent of motion granularity and variety is CSI capable of distinguishing in practice. 3.3 Oneleapfurther: WiFiradar Over the past five years, CSI has spawned various applications and its application scenarios continue to expand. As an upgrade for RSS, it is natural to improve performance of some applications simply by replacing RSS with CSI. CSI also enables various applications infeasible with RSS alone, such as gesture recognition, breath detection, and complex environment sensing. Nevertheless, CSI is not a panacea, and its improvement in sensing granularity is still incomparable with radar signals. Some envisioned