Research On Context-aware Techniques Based On Channel State Information

Context-aware techniques originated from ubiquitous computing enable computing devices to sense surrounding environment through sensors and data analysis and processing techniques.The environment mentioned above is not limited to physical environment,but also location,state,and even emotion,body language and gesture information.Context-aware techniques have been widely used in people's daily life,such as human intrusion detection based on localization,human-computer interaction and smart home based on activity recognition,which provides us with more intelligent services.Recently,with the development of wireless communication technology,wireless signals have expanded from a traditional communication medium to a platform for wireless sensing,and ubiquitous WiFi signals make WiFi-based context-aware techniques feasible.However,traditional WiFi-based techniques mostly utilize the Received Signal Strength Indication(RSSI)that is a coarsegrained description of signals and fluctuates with time,as a result,limiting its recognition accuracy.Hence,there is a need for a new context-aware technology based on WiFi.This thesis aims to study the context-aware techniques based on fine-grained Channel State Information(CSI).We make a classification for context-aware techniques combining with its applications and related technologies and analyze and compare RSSI and CSI.Considering of the disadvantages of RSSI-based context-aware techniques,we study and design three CSI-based systems:(1)human localization system,which uses CSI characteristic that it stays stable in the same propagation environment while shows different characteristics in different environments to identify the person is indoors or outdoors;(2)exercise activity recognition system,which extracts CSI waveform that contains both time and frequency information as activity feature and devises a series of denoising methods(i.e.,low-pass filtering,PCA and median filtering)to eliminate noise,effectively realizing activity recognition and quality evaluation;(3)gesture recognition system,in which an improved Linear Discriminant Analysis algorithm(I-LDA)is designed to reduce the dimension of multidimensional behavior signals,lowering computational cost.Additionally,behaviors are learned by Logistic Regression Algorithm(LRA)where bandwidth ratios in energy spectrum are selected as features to eliminate the impact of different speeds.Finally,the systems mentioned above are implemented on commercial wireless devices and its performances are verified and evaluated.Experimental results show that our systems are feasible and robust.