Performance Enhancement of Wi-Fi Fingerprinting-based Indoor Positioning using Truncated Singular Value Decomposition and LSTM Model

Wi-Fi based indoor positioning has been considered as the most promising approach for civil location-based service due to the widespread availability Wi-Fi systems in many buildings. One of the most favorable approaches is to employ received signal strength indicator (RSSI) of Wi-Fi access points as the signals for estimating the mobile object locations. However, developing a solution to obtain high positioning accuracy while reducing system complexity using traditional methods as well as deep learning based methods is still a very challenging task. This paper presents a proposal to combine the Truncated Singular Value Decomposition (SVD) technique with a Long Short -Term Memory (LSTM) model to enhance the performance of indoor positioning system. Experimental results on a public dataset demonstrate that the proposed approach outperforms other state-of-the-art solutions by means of positioning accuracy as well as computational cost.