Multipath Mitigation for Indoor Positioning with Empirical Mode Decomposition and Spatial Autoregressive Neural Network Methods

Abstract

The multipath in the wireless channel leads to a bias of time of flight (TOF) observations, which degrades indoor positioning accuracy and performance. This paper proposes a method for modeling ranging errors to mitigate multipath effects and using a filtering algorithm in signal pre-processing. By combining traditional spatial interpolation methods with neural networks, we propose a spatial autoregressive network architecture, which fits the nonlinear relationship between the spatial distance and the spatial autocorrelation weight to calculate the error. The spatial autoregressive neural network (SANN) overcomes the limitation of traditional interpolation methods when calculating spatial weights and the disadvantage of existing methods that directly fit errors by spatial distances, which neglects the spatial correlation of errors, resulting in high prediction accuracy. This paper also uses empirical modal decomposition (EMD) to pre-process the signal to reduce the noise in the signal. The lab experiments demonstrate that EMD effectively minimizes signal noise and reduces raw data variance by 30%. SANN effectively mitigates the multipath effects. The experimental results show that the average positioning accuracy achieves 9.2 cm. Even in dynamic tests, high accuracy and robustness results are displayed.