A Machine Learning Multipath Mitigation Approach for Opportunistic Navigation with 5G Signals

Mohamad Orabi, Ali A. Abdallah, Joe Khalife, and Zaher M. Kassas

Peer Reviewed

Abstract:	The ability of different neural networks to mitigate multipath signals for opportunistic navigation with downlink 5G signals is assessed. Two neural networks, namely feed-forward neural networks (FFNNs) and time-delay neural networks (TDNNs), are designed to learn multipath-induced errors on a 5G receiver’s code phase estimate. The neural networks use inputs from the autocorrelation function (ACF) to learn the errors in the code phase estimate of a conventional delay-locked loop (DLL). A ray tracing algorithm is used to produce high fidelity training data that could model the dynamics between the line of sight (LOS) component and the non-line of sight (NLOS) components. Cross-validation methods are used on FFNNs to examine the sensitivity of the out-of-sample error on the number of hidden layers, number of neurons per layer, and regularization constant that limits the complexity of the hypothesis space. Moreover, TDNNs with varying access to the time history of the ACF taps are assessed. Experimental results in a multipath-rich environment are presented demonstrating that the proposed TDNN achieved ranging root-mean squared error (RMSE) reduction of 27.1% compared to a conventional DLL.
Published in:	Proceedings of the 34th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2021) September 20 - 24, 2021 Union Station Hotel St. Louis, Missouri
Pages:	2895 - 2909
Cite this article:	Orabi, Mohamad, Abdallah, Ali A., Khalife, Joe, Kassas, Zaher M., "A Machine Learning Multipath Mitigation Approach for Opportunistic Navigation with 5G Signals," Proceedings of the 34th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2021), St. Louis, Missouri, September 2021, pp. 2895-2909. https://doi.org/10.33012/2021.17990
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In