CSI-fingerprinting Based Human Indoor Localization in Noisy Environment using Time-Invariant CNN

Indoor localization of humans in practical situations has long been a challenging task due to limitations of cost and available device numbers. CSI-fingerprinting method, which predicts user’s coordinates based on Wi-Fi signals, are recently considered as a promising approach for low-cost and device-efficient localization. However, such method suffers from low prediction accuracy in noisy environments. The noisy CSI data are known to randomly fluctuate through time, which makes it difficult for existing models to capture meaningful features for accurate prediction. This research addresses this problem by proposing Time-Invariant Convolution Block. The proposed block ignores the order of CSI data, which is known to contain no information, but rather focuses on the frequency of CSI data. In the experiment, this module showed high prediction accuracy and stability compared to conventional methods.