Abstract
Fingerprinting Localization, a common technique used in indoor positioning uses short-range radio frequency and addresses problems with multi-path. Although, there are challenges in fingerprinting approaches like spatial ambiguity, long distances, low-bandwidth, scalability, cost and size constraints. Additionally, LoRa is more efficient in comparison to BLE and Wi-Fi considering challenges in GPS based applications. In offline phase of fingerprinting, 2D data at different locations have been collected in a university building, gathering RSSI values from the gateways at a fixed location. The online phase estimates the mean location error using Deep Learning models. Indoor experiments using DL techniques achieve 1.2-2.0 [m] of mean distance error. On the contrary, deep-learning techniques were implemented for publicly available outdoor data-source collected from several LoRa WAN base-stations and nodes from Antwerp city, Belgium. Interpolation techniques using denoising auto-encoders have helped to interpolate outliers for this data. Our results have demonstrated that a mean distance error of 191.52 [m] from LSTM has out-performed results from the KNN algorithm. Impact of several LSTM design parameters have helped us to understand back-propagation algorithm in LSTM, to improve our results.