A UWB-Based Indoor Positioning System Employing Neural Networks Binghao Li 1 & Kai Zhao 1 & Eduardo Benitez Sandoval 2 # Springer Nature Switzerland AG 2020 Abstract Because of its high time-domain resolution, ultra-wideband (UWB) technology is capable of precise range measurement and has been applied in many accurate indoor positioning systems. This paper presents a practical indoor positioning and navigation system using the UWB modules we developed based on Decawave UWB chipset. In order to enhance the performance of the system in NLOS scenarios, a neural network approach is integrated. The feasibility of the combined approach is analysed and its accuracy is evaluated. The experiment results indicate that the neural network can maintain a stable accuracy no matter whether it is in an NLOS scenario and it is capable to identify the locations on two trajectories which are next to each other with very high confidence. Keywords Ultra-wideband (UWB) . Indoor positioning . Neural network Introduction Indoor positioning has attracted attention from both the aca- demic world and the industry for about two decades. Technologies including bluetooth low energy (BLE), Wi-Fi, ultra-sound, laser and ultra-wideband (UWB) have been ap- plied in indoor positioning systems. However, there is still no perfect solution for indoor positioning as a practical indoor positioning system must balance the accuracy, the labour cost, the facility expense, the reliability and the complexity. No technology can meet the requirements in all these aspects. Generally speaking, UWB devices can achieve decimeter- level accuracy, which means it can satisfy the requirements of most indoor positioning applications with regards to the accuracy. The range of UWB devices can be over 100 m (un- der certain conditions) and the positioning process only re- quires the coordination of the anchors. Therefore, the deploy- ment and maintenance of a UWB positioning system are not labour-costly. However, the cost of UWB devices is higher compared with the costs of BLE and Wi-Fi devices. UWB has drawn a lot of attention since Decawave released its UWB chip in 2012. The Decawave’s products are capable of two-way ranging (TWR) (APS013 n.d.) which enables the active positioning on the tag. The price of a Decawave UWB chip can be as low as 8USD (DecaWave n.d.) and its accuracy is better than other UWB devices like Ubisense and BeSpoon (Jimnez Ruiz and Granja 2017). Given these advantages of the UWB positioning system, many experiments that analyse the performance of UWB ranging and positioning have been done. Most of them men- tioned the non-line-of-sight (NLOS) scenario and discussed the solution to improve the accuracy under certain circum- stances. Unlike the infrared and ultrasound sensors, UWB device can work in NLOS scenario although its accuracy de- clines (Alarifi et al. 2016). NLOS can significantly cripple the signal on the direct path. It is possible to detect the NLOS by analysing the features of the received waveform (Guvenc et al. 2007; Decarli et al. 2010; Schroeder et al. 2007). It is reported that detecting NLOS signals can help to adopt proper filter on the signals and improve the accuracy in a static experiment (Zhang et al. 2016). However, these algorithms can only iden- tify the NLOS signals but not correct the error. More impor- tantly, in a complex environment, the NLOS can be caused by a wall, two walls or any other types of obstacles and both the features and the effect to the ranging results can vary. The algorithms in (Guvenc et al. 2007; Decarli et al. 2010; Schroeder et al. 2007; Zhang et al. 2016) generally only clas- sify the signals into NLOS and line-of-sight (LOS) scenarios. No more information for range correction is provided. * Binghao Li binghao.li@unsw.edu.au 1 School of Minerals and Energy Resources Engineering, University of New South Wales, Sydney, Australia 2 Faculty of Art and Design, University of New South Wales, Sydney, Australia Journal of Geovisualization and Spatial Analysis (2020) 4:18 https://doi.org/10.1007/s41651-020-00059-2