RSSI fingerprinting for Localization using Low Power Wide Area Network: Deep Regression Tuning

Internet of things has caused a myriad of devices is connected to the network in everywhere. In recent years, numerous devices have been connected to a large communication network. In order to implement this connection, Low Power Wide Area Network technologies such as Sigfox and LoRaWAN have been developed. Most LPWAN IoT applications utilize location information which it can be done by GPS. Since this common method uses more energy in LPWAN, it can be used energy-efficient solutions such a Received Signal Strength (RSS)-based fingerprinting localization. In this research, there openly available fingerprinting datasets have been used to estimate location. These are including rural and urban Sigfox and urban LoRaWAN datasets. A Deep Neural Network algorithm has been applied to the datasets. With the appropriate tuning of hyper-parameters, the achieved mean square localization error was 1.94 for rural Sigfox dataset, 1.33 for urban Sigfox and 0.2 for LoRaWAN dataset.

Keywords:

Fingerprinting , Localization , LPWAN , Deep Learning