Research On Indoor Positioning Technology Of Dual Bands WiFi Based On Capsule Network

In recent years,various high-tech technologies have emerged with the development of artificial intelligence,and indoor positioning technology has become more and more popular.It has been widely applied and promoted.It not only facilitates people’s lives but also creates enormous economic value.However,during the propagation of the WiFi signal,it is easy to be interfered with obstacles,and the signal fluctuation is significant which resulting in low accuracy of positioning.To overcome these problems,we reduce the influence of environmental factors firstly.Then the positioning accuracy is improved by using the SVM model to distinguish the NLOS or LOS environment and employing the capsule networks to derive the users’ positions with the WiFi2.4G and 5G signals.In this paper,the characteristics of dual bands WiFi signal are analyzed in detail,and it is concluded that the WiFi2.4G signal has a large fluctuation range in the LOS state but good signal penetration in the NLOS state.On the contrary,the fluctuation range of the WiFi5G signal is small and stable in the LOS state,and the signal penetration is poor in the NLOS state.Based on the above conclusions,this paper proposes a fuse learning method that uses the advantages of the two signals in different states.Besides,this paper also proposed a positioning system and our system named as Bi CN method(Bi-modal capsule network for indoor localization using commodity WiFi devices).We applied the capsule network to the indoor positioning field for the first time and trained a capsule network model with WiFi2.4G and WiFi5G signals respectively.We can obtain two predicted positions simultaneously in the positioning phase,and use the weighting algorithm to obtain the final predicted position.To verify that the positioning system proposed in this paper has high precision and good robustness,the positioning accuracy of common indoor positioning algorithms is compared under the field environment and the simulated environment.The experimental results show that the proposed positioning method has the highest accuracy and good robustness.At the same time,we tested the positioning effect of the system trained by single signal data and found our proposed BiCN method using WiFi2.4G and WiFi5G to locate the position can speed up the convergence.