The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLVI-3/W1-2022
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLVI-3/W1-2022, 61–66, 2022
https://doi.org/10.5194/isprs-archives-XLVI-3-W1-2022-61-2022
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLVI-3/W1-2022, 61–66, 2022
https://doi.org/10.5194/isprs-archives-XLVI-3-W1-2022-61-2022
 
22 Apr 2022
22 Apr 2022

CONVOLUTIONAL NEURAL NETWORKS BASED GNSS SIGNAL CLASSIFICATION USING CORRELATOR-LEVEL MEASUREMENTS

C. Jiang1, Y. Chen1, B. Xu2, J. Jia1, H. Sun1, Z. He3, T. Wang4, and J. Hyyppä1 C. Jiang et al.
  • 1Department of Photogrammetry and Remote Sensing, Finnish Geospatial Research Institute, Masala, FI-02430, Finland
  • 2Interdisciplinary Division of Aeronautical and Aviation Engineering, The Hong Kong Polytechnic University, Hong Kong
  • 3Department of Geomatics Engineering, University of Calgary, Canada
  • 4Huawei Helsinki Research Centre, Finland

Keywords: GNSS, NLOS, Multipath, Convolutional Neural Networks

Abstract. In urban areas, the None-Line-Of-Sight (NLOS) and Multipath (MP) signals are the major issues degrading the GNSS position accuracy. Signal reception type should be identified before correcting the NLOS or MP induced errors. Signal features, i.e., signal strength, change rate of received signal strength, difference between delta pseudo-range and pseudo-range rate, have been explored in signal reception type classification. In this letter, with the aim to improve the signal classification accuracy, we propose a new GNSS NLOS/MP/LOS signals classification method using the correlator-level measurements. Firstly, vector tracking (VT) is employed to generate correlator-level measurements; secondly, a deep learning method, Convolutional Neural Network (CNN), is employed to automatically extract the features and identify the signal reception type, correlators’ outputs calculated at different code phases are employed as the inputs of the CNN. Field test is carried out for assessing the performance of the proposed method, and the CNN method obtains state-of-art performance compared with the K-nearest Neighbors Algorithm (kNN) and Support Vector Machine (SVM) methods.