Unsupervised Learning Based Acoustic NLOS Identification For Smartphone Indoor Localization

Compared with the existing indoor localization technologies,acoustic localization technology has the advantages of high localization accuracy,low cost and a broad applicate prospect.In the complex indoor environment,the non-line-of-sight propagation of acoustic signal seriously affects the positioning accuracy of the system.Through NLOS identification and processing,the localization accuracy and stability can be improved.Therefore,this thesis studies the acoustic signal NLOS identification method based on unsupervised learning.The main contents of this thesis are as follows:(1)An acoustic channel parameter estimation method based on hyperbolic frequency modulation signal is proposed.By analyzing the influence of various environmental parameters on the acoustic channel parameters,the gain-delay estimation problem of acoustic channel is transformed into the relative gain-delay estimation problem.The Doppler frequency shift invariant property of hyperbolic FM signal is used to eliminate the Doppler frequency shift term,and the relative gain-delay of acoustic channel is estimated accurately.The experiment results show that when the SNR is [-15 dB,15 dB],the relative gain error ratio of the hyperbolic FM signal is less than 10%,the relative delay error ratio is less than two sampling points,and the relative gain-delay error ratio is significantly lower than that of the linear FM signal.(2)A NLOS identification strategy without prior information is proposed.Based on the unsupervised learning method after parameter optimization,two clusters without tag information are obtained and the features of the two clusters are extracted.Bayesian decision method is used to obtain the tag information of clusters to realize NLOS identification of acoustic signals without prior information.The experimental results in three scenarios show that the accuracy of Bayesian decision method is not less than 90%,the average accuracy is 94.43%.