Neural Network-Based Multipath Mitigation Method for Precise Indoor Positioning

Min-Ji Kim, Ki-Hyun Kim, Changdon Kee, O-Jong Kim

Abstract:	Multipath, a significant cause of errors in radio navigation, results from signal reflection at objects and structures. While most errors can be addressed through mathematical methods, multipath remains challenging. This problem becomes even more challenging in complex indoor environments. This study focuses on multipath detection of single-transmitter-based positioning in an indoor environment. A single-transmitter-based positioning system, utilizing an antenna array and suitable for indoor navigation, reduces the need for multiple transmitters where satellite navigation isn't feasible. This system involves a transmitter that creates multichannel signals and transmits them using an antenna array. It utilizes carrier phase measurements to achieve precise positioning and to attain minimal levels of noise. According to previous studies on this type of system, most measurement errors are caused by multipath. Thus, multipath mitigation is vital for enhanced accuracy. Due to the difficulty in accurately modeling numerous signals reflected within indoor environments, this study proposes a machine learning-based model for detecting and categorizing signals distorted by multipath. The method employs two time-frequency analysis techniques. One of them is the Continuous Wavelet Transform (CWT), which generates a scalogram representing the absolute values of the signal's CWT coefficients. The other is the Short-Time Fourier Transform (STFT), producing a spectrogram that represents the magnitude of each frequency component. The transformed CWT and STFT values contribute to supervised learning. Neural networks are used for multipath detection and have employed various architectures through an empirical optimization process, considering time-frequency analysis methods and desired output. Carrier phase-based multipath detection is carried out using neural networks, followed by position calculation to mitigate multipath errors. The experiments were conducted in an indoor environment consisting of steel frames and concrete walls, with various objects present. After the multipath mitigation based on this method, about 30% of the positioning error was reduced.
Published in:	Proceedings of the 36th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2023) September 11 - 15, 2023 Hyatt Regency Denver Denver, Colorado
Pages:	1999 - 2012
Cite this article:	Kim, Min-Ji, Kim, Ki-Hyun, Kee, Changdon, Kim, O-Jong, "Neural Network-Based Multipath Mitigation Method for Precise Indoor Positioning," Proceedings of the 36th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2023), Denver, Colorado, September 2023, pp. 1999-2012. https://doi.org/10.33012/2023.19333
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In