Deep Learning of GNSS Signal Correlation

This paper studies the use of deep learning models to model the complex effects of multipath propagation on GNSS correlation outputs. Particularly, we aim at substituting standard correlation schemes (that are optimal under nominal conditions) with Neutral Network (NN)-based correlation schemes, that are able to learn multipath channels, otherwise challenging to be captured by physics-based models. The solution involves the use of a Deep Neural Network (DNN) structure, which exploits the optimality of standard correlation schemes at the same time that adjusts its behavior to better model multipath. The proposed solution can be used in substitution of standard correlation, then seamlessly plugged on acquisition and tracking receiver blocks. Results on the currently trained model show promising time-delay tracking performances, when compared to standard correlation processing.