| Recently,with the popularization of wireless networks,WiFi devices have come out in many families.With the development of wireless sensing technology,WiFi can be used for localization,identification,and movement recognition by analyzing the influence on WiFi signals from the target.AOA(Angle of Arrival)estimation has attracted lots of attention as the basic technique of wireless sensing.Although wireless sensing has made great progress,limited to the design of commercial devices and the cost of hardware,there is still a problem that how to realize low-cost and high-accuracy AOA estimation with commercial WiFi devices.This thesis is focused on the problems mentioned above and provides some ideas for AOA estimation problems that can realize wireless localization.Firstly,the antenna array of commercial WiFi devices has changed from linear array to different shapes,such as circular array and square array.But most AOA estimation algorithm is based on the linear array model and is hard to generate a unitive AOA estimation model for the irregular and nonuniform array which would affect the capability of AOA estimation.For all mentioned above,this thesis proposes an AOA estimation method with irregular array based on the deep-learning network.This method utilizes the phase information as the main feature.It also uses the directional antenna to establish the relationship between the direction and amplitude of channel state information which becomes the auxiliary feature.Then,with the capability of data mining of the deep-learning network,it solves the problem of establishing a unitive model for irregular arrays and realizes AOA estimation with irregular arrays.Secondly,most commercial WiFi devices only have a few antennas.However,the accuracy of AOA estimation is correlated to the number of antennas,which makes commercial WiFi can’t achieve high-accuracy AOA estimation.To solve the problem,this thesis proposes a multi-view-based AOA estimation method.Specifically,this method utilized the characteristic that the resolution of AOA estimation is different in different views and proposes a multi-view idea which can utilize AOA information in the space domain and frequency domain.This method also establishes a multi-view fusion model which can fuse these views.It converts the AOA estimation problem to a solvable optimizing problem,so it can avoid the interference and improve the accuracy.With the method,AOA estimation accuracy can be improved without extra hardware cost.Finally,to verify the methods mentioned above,this thesis proposes an AOA estimation experiment with irregular arrays and an AOA estimation experiment that uses the multi-view AOA estimation method.These experiments verify the AOA estimation method with irregular array based on the deep-learning network and the multi-view AOA estimation method,which can realize high-accuracy AOA estimation with irregular array and realize low-cost and high-accuracy AOA estimation,separately. |
