|Abstract:||Deep neural networks applied to GNSS positioning are promising for accurate positioning in diverse environments by leveraging data-driven models of measurements and position. However, creating and deploying neural networks in practical positioning applications is challenging due to the large data, training time and memory requirements. In this work, we propose a novel hybrid learning-based approach for GNSS positioning that builds on traditional positioning models for improved data and memory efficiency. Our approach augments the models of the geometry matrix and the expected pseudorange measurement with data-driven components to improve the positioning performance in scenarios that deviate from the modeling assumptions. Through experiments on simulated data with diverse measurement errors, we show that our proposed approach achieves low positioning errors compared to other model-based and neural network approaches with little training while requiring lesser memory for evaluation.|
Proceedings of the 35th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2022)
September 19 - 23, 2022
Hyatt Regency Denver
|Pages:||1209 - 1219|
|Cite this article:||
Gupta, Shubh, Kanhere, Ashwin V., Shetty, Akshay, Gao, Grace, "Designing Deep Neural Networks for Sequential GNSS Positioning," Proceedings of the 35th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2022), Denver, Colorado, September 2022, pp. 1209-1219.
ION Members/Non-Members: 1 Download Credit