3D Convolutional Neural Network-Aided Indoor Positioning Based on Fingerprints of BLE RSSI

This paper deals with an indoor positioning via deep learning techniques based on the received signal strength indication (RSSI) of Bluetooth low energy (BLE) beacon signals. In fingerprint positioning, a site-survey is conducted in advance to build the radio map, which can be used to match radio signatures with specific locations. It takes into account the complex effects of real-environments and enables highly accurate indoor positioning. However, even in static indoor environments, the observed RSSI values are statistically fluctuated due to random wireless channels, leading to severe performance degradation of the fingerprint estimation. To address this issue, we introduce the three-dimensional convolutional neural network (3D-CNN) to fingerprint positioning with the RSSI data set (available as big data). The 3D-CNN can handle 3D spatiotemporal structures of RSSI data set and utilize the temporal fluctuations that finger-print cannot capture to enhance the positioning accuracy. The experimental results show the validity of our proposed scheme using the 3D-CNN-based fingerprint positioning, as compared to the typical positioning schemes on the basis of the feed-forward NN (FNN) and two-dimensional CNN (2D-CNN).