A Deep Learning Approach for the Classification of Multipath Ranging Errors in Challenging Urban Environments

Distortion to the correlation function caused by multipath and non-line-of-sight signals can result in pseudorange errors on the order of several tens of meters in urban canyon environments. To address this problem, a deep learning approach for classifying multipath ranging error from a global navigation satellite systems (GNSS) receiver correlation function is presented. This approach uses a one-dimensional convolutional neural network, suitable for embedded applications, to classify the magnitude of pseudorange error associated with correlation functions. The network is trained and tested on live GNSS data collected in a challenging urban environment, and the capability of the model to remove high error measurements for a least-squares positioning solution is explored. The network has proven to be effective at detecting measurements with high multipath ranging error, and the removal of detected measurements reduced positioning error by up to 80%.