A new deep learning-based distance and position estimation model for range-based indoor localization systems

Many fine-grained indoor localization systems rely on accurate distance estimation between anchors and a target node to determine its exact position. The Received Signal Strength Indicator (RSSI) is commonly used for distance estimation because it is available in most low cost standard wireless devices. Despite the cost efficiency, the distance estimation accuracy in the RSSI-based ranging model needs to be enhanced, especially indoors. The RSSI is sensitive to multiple indoor factors that fluctuate in time and space and lead therefore to its variation. These factors are the origin of the distance estimation error increase in RSSI-based ranging models which in turn raise the position estimation error. Previous works have presented different in-site self calibration processes to improve the accuracy of distance estimation using RSSI-to-distance samples. It permits to settle the parameters of the RSSI-based ranging model such as the Path Loss Model (PLM) and to mitigate the changing behavior of the RSSI. However, the RSSI measurement depends not only on the distance between the transmitter and the receiver but also on the indoor ambient temperature and humidity variations. Besides, indoor obstacles such as furniture, metallic surfaces or walls have also an impact on the RSSI measurements. We present in this paper a new RSSI-based indoor ranging model using deep learning on collected in-site samples to ensure efficient and autonomic calibration process. This permits to mitigate disturbing factors such as temperature, humidity and noise in order to increase the accuracy of both distance and position estimations. The experimental results have shown that our ranging model has improved not only the precision of distance estimation but also the position estimation in the range-based indoor localization systems.