|Abstract:||In this work, we propose a convolutional neural network (CNN) model for the detection of multipath in Global Navigation Satellite System (GNSS) signals. The main idea is to represent the signal as multiple 2D data grids forming images in the time-frequency domain. The images are then processed by the convolutional layers to extract relevant features. As labelled multipath data are very scarce, we have also developed a synthetic correlator output generator with several tuning strategies for the multipath scenario based on doppler shift, propagation delay and phase estimation errors. In order to assess the overall performance of the technique and validate the possibility to embed it in GNSS receivers, after training with synthetic data, the prediction model has been implemented on a low power vision processing unit (VPU) in the form of a USB dongle which is characterized by a low power consumption envelope in the order of 1W. The results have shown that the model can detect multipath with good accuracy at a faster rate than GNSS receivers sample processing rate on tracking stage.|
Proceedings of the 33rd International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2020)
September 21 - 25, 2020
|Pages:||2018 - 2029|
|Cite this article:||
Munin, Evgenii, Blais, Antoine, Couellan, Nicolas, "GNSS Multipath Detection Using Embedded Deep CNN on Intel® Neural Compute Stick," Proceedings of the 33rd International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2020), September 2020, pp. 2018-2029.
ION Members/Non-Members: 1 Download Credit