Single-Site Indoor Fingerprint Localization Based on MIMO-CSI

Jiancun Fan,Jianxiong Zhang,Xiaoyuan Dou

DOI: https://doi.org/10.23919/jcc.2021.08.014

2021-01-01

China Communications

Abstract:In order to achieve higher accuracy and lower cost of indoor localization, we propose a positioning method using multiple input and multiple output (MIMO) channel state information (CSI) as a fingerprint. The method can be divided into three stages, feature extraction, offline training and online localization. In the feature extraction, the segmented average and principal component analysis (PCA) are used to reduce the data dimension and decrease system complexity. In the offline training, the deep neural network (NN) model is trained to implement the position classification. In the online localization, the data are input into the trained NN model first, and then its output is further processed by weighted k-nearest neighbor (WKNN) technology to estimate the position. Experimental results show that the proposed method can significantly reduce the positioning error compared to other methods and the average error is 1.39m in a complex indoor environment.