Research On CSI Passive Indoor Positioning Method Based On Keras

In recent years,with the widespread popularity of mobile devices and the rapid development of mobile Internet,many mobile applications include a positioning module,which has brought more and more convenience to people's lives.At present,in outdoor environments,the Global Positioning System(GPS),Beidou satellite navigation system and Galileo satellite navigation system,this positioning technology based on satellite signals is mature,good real-time,strong anti-interference ability,high positioning accuracy,can meet the people Daily life needs.However,due to the complex and changeable indoor environment,their positioning accuracy in the indoor environment is relatively low,so they are not suitable for indoor positioning research.However,in some special occasions,indoor positioning also has a strong demand.At present,the indoor positioning system based on channel state information(Channel State Information,CSI)is widely concerned and applied by researchers because of its simple operation,no need for users to carry portable equipment,and high positioning accuracy.This paper proposes a passive indoor positioning method for channel state information based on Keras deep learning framework.The main work is as follows:We analyzed the transmission characteristics of wireless signals,explained in detail the transmission methods and composition of wireless signals,and summarized various factors that affect signal transmission.Secondly,it introduces some traditional indoor positioning methods,and points out their advantages and disadvantages,which mainly introduces indoor positioning methods based on RSSI and CSI.On the basis of the above,an indoor positioning system based on CSI is proposed and introduced in detail.First,we extracted amplitude information and phase information from the collected CSI data,and preprocessed them,removed the noise by Discrete Wavelet Transformation(DWT),and adopted Principal Components Analysis(PCA))Data dimensionality reduction,mapping high-dimensional CSI data to low-dimensional space.After that,we perform feature selection on the data,that is,divide the pre-processed amplitude data and phase data in proportion to form a new data set.Finally,a Long Short-Term Memory(LSTM)based on Keras deep learning framework is used to train the data,and a training model based on amplitude and phase is obtained.In the online phase,the training model performs regression prediction on the position to be measured by calling a prediction function to obtain thecoordinates of the position to be measured.Experiments were conducted in an open environment and a complex laboratory environment to evaluate the performance of the method,and compared with the current most advanced indoor positioning solutions.At the same time,a variety of comparative experiments were conducted to verify our method.