Deep Learning Coarse Doppler Estimation for LEO Signals of Opportunity

This article presents the development of a deep neural network (DNN) model based on the de-noising autoencoder (DAE) architecture, which is trained with the objective of detecting and estimating the coarse Doppler of signals of opportunity (SOOP) from low-orbit (LEO) communications satellites. In particular, the work is focused on LEO signals from the Orbcomm constellation, and scenarios where interfering signals with much higher power spectral density (PSD) are present. The model is trained to first ignore the noise and the interferences, and subsequently estimate the coarse Doppler of the Orbcomm signal of interest. The resulting performance confirms the feasibility of using DNN-based models for opportunistically estimating the Doppler of LEO signals. This facilitates the integration of LEO and GNSS observables, thus providing an additional layer for improving the GNSS positioning performance in harsh working conditions.