CS-Dict: Accurate Indoor Localization with CSI Selective Amplitude and Phase Based Regularized Dictionary Learning

With the increasingly growing demand for indoor location-based services(LBS) in the field of wireless sensing, Wi-Fi have been the mainstream method in indoor localization for the reasons of easy deployment and the popularity of signal. Channel State Information (CSI) is extracted from the physical layer of WiFi network interface cards and includes more fine-grained signal characteristics than received signal strength index (RSSI) which is commonly used in the literature. In this paper, we propose CSI selective dictionary (CS-Dict), an accurate model-free indoor localization algorithm using only one access point simultaneously. CS-Dict mainly contains two parts: CSI feature enhancement and over-complete dictionary learning. In the feature enhancement, CSI features with high reliability are selected as the input for dictionary learning. In the over-complete dictionary learning, we utilize the regularized K-SVD to perform a dictionary representation of selective CSI features in each reference point. Finally, a similarity measurement between the real-time measured CSI and the learned dictionary is performed to find the best match for position estimation. An extensive experiment is deployed in two typical indoor environments, the results show that the mean error are 0.12 m and 0.23 m respectively.