A deep learning-based indoor localization algorithm using CSI measurement and time series decomposition

Abstract

Recently, indoor localization techniques have received considerable attention due to their wide utilization. In this paper, a deep learning based indoor localization algorithm using channel state information (CSI) measurements and time series decomposition has been proposed. First, the mean deviation based signal preprocessing is proposed to mitigate the abnormal data of CSI original amplitude information. And then the empirical wavelet transform (EWT) and variational mode decomposition (VMD) are proposed to extract the features. Next, the relative position matrix is proposed to obtain the CSI EWT feature image and CSI VMD feature image. At last, the dual-stream convolutional neural network (CNN) is designed for offline training by the middle layer fusion. Experimental results show that the proposed algorithm has better localization performance than the chosen localization approaches.