Research On Wireless Channel Classification Based On Indoor Wireless Local Area Network

Wireless local area network is widely used in indoor positioning systems due to its high data rate,low power consumption and low overhead,However,there are many difficulties resulted by narrow scene,obstacles and non-line of sight propagation in indoor positioning.In recent years,with the vigorous development of artificial intelligence,studies have shown that the combination of machine learning algorithm and NLOS recognition has greatly improved the performance of NLOS recognition compared with traditional algorithms.However,NLOS recognition still faces many problems: the wireless channel distribution has no general law under LOS/NLOS propagation;the classification algorithm,the selection of algorithm parameters and the extraction of feature data are all problems faced by line-of-sight/non-line-of-sight recognition.This paper improves the line-of-sight/non-line-of-sight classification algorithm and feature extraction,which improves the recognition performance from these situations.The research is mainly divided into two parts:1.This paper proposes a classification algorithm based on twin support vector machines and improved particle swarm optimization for line-of-sight/non-line-of-sight propagation recognition.At present,there are some problems such as poor performance and lack of optimization of algorithm parameters in the classification of non-line-of-sight propagation.Firstly,the TWSVM algorithm is used to replace the traditional SVM algorithm for classification experiment,which reduces the algorithm complexity and improves the algorithm performance.Then,a PSO-WF algorithm is proposed to replace the traditional PSO algorithm to optimize the parameters of the classification algorithm,which solves the problems of slow convergence speed and easy to fall into local search in THE PSO algorithm.The simulation results show that the TWSVM algorithm has better classification performance than SVM algorithm,and PSO-WF combined with TWSVM algorithm has better classification effect than TWSVM algorithm and PSO-TWSVM algorithm alone.2.In this paper,doppler power spectrum data is used as the characteristic data of channel scene division to solve the problem of poor performance of channel scene division in wireless LAN positioning system.In view of the existing problems of channel scene classification,such as limited bandwidth of indoor positioning system,poor performance of time domain feature classification,this paper proposes to use Doppler power spectrum data for scene classification.Three different scenarios were constructed through ray tracing for channel modeling and simulation,and the impulse response data of wireless communication in three simulation scenarios were processed,and the Doppler power spectrum data and time domain characteristics of different scenarios were extracted.Finally,the performance was compared by K-means,DBSCAN and GMM clustering algorithms.The simulation results show that although the clustering performance of Doppler power spectrum data is less than that of the high time-delay resolution feature data,the time-domain feature data in different channel scenarios overlap greatly after the communication bandwidth is reduced,and the Doppler power spectrum data has better clustering performance than time-domain feature data.