An Ephemeris for Cruisable UAV Pseudolite Navigation System Based on Deep Learning Networks

When a pseudolite is used to provide positioning services to users, it needs to provide position information by broadcasting ephemeris parameters. As part of resilient positioning, navigation, and timing (PNT), a UAV pseudolite system can serve users independently as a GNSS backup system, but unlike satellites that move periodically, the motion of UAVs is often variable, so it brings difficulties for how to make accurate predictions of the UAV's trajectory in a future segment within the ephemeris parameter update period. Therefore, based on a cruise UAV, we proposed an ephemeris based on a prediction method combined with convolutional neural network (CNN) and long short-term memory (LSTM) networks. The true position is divided into positions on the intended route and their deviations by mapping, and we use the motion model, the Singer model with Kalman filter, and the CNN-LSTM networks to predict the position. Moreover, the network updates are achieved by a network monitoring loop, and adaptive adjustment of the parameter update period is achieved by a prediction accuracy monitoring loop. Moreover, the ephemeris parameters are obtained by fitting the prediction results. The experimental results verify the performance of the designed ephemeris, the root mean squared error (RMSE) value of ephemeris parameter characterization error is less than 5 m, and the average occupied bits is about 35 bits; even using the sample data of UAV maneuvering at extreme speed with poor control performance, the RMSE of ephemeris parameter characterization error is 6.56 m, and the average occupied bits is 69.0 bits.