GNSS Satellite orbit predictions are used to reduce the Time to First Fix (TTFF) of a satellite positioning device.

The accuracy of GNSS-based navigation is determined by GNSS raw data quality (pseudo-range, doppler, carrier phase), navigation data, multipath, ionosphere and so forth. In some strict and urban environments, the SNR (signal-to-noise ratio) of GNSS signals is weaker and around 25 dBHz or less, thus it is not realistic to download ephemeris data broadcasted from Satellites. Not only TTFF, but also the accuracy of GNSS positioning relies on the accuracy of predicted orbit/clock data.

The satellite prediction is divided into two parts, namely orbit prediction and clock prediction. The prediction is based on precise orbit data (SP3) and Earth Orientation Parameters (EOP).

In the satellite clock prediction, LSTM (Long Short-Term Memory) is selected as a priori model to verify the availability of time sequence model. Since satellite clock data are unstable, only short period data are used for training process, which is highly affected by the quality of data. To improve the quality of data, a data preprocessing is designed to cover unreliable data, such as fluctuation, clock reset and data loss.

The strategy is combining LSTM and linear regression. Three functions are designed to fit the curve of satellite clock bias. The functionality of three functions are main curve, noise, and boundary.

Airoha Technology (MediaTek’s subsidiary) integrated the deep learning-based method (LSTM) for improving the accuracy and robustness in next generation extended GNSS satellite orbit/clock prediction technology. To improve the GNSS positioning accuracy and stability, especially challenges encountered in wearable use cases: weaker signal and poor efficiency in antenna design with limited spaces.

Learn more on https://www.ion.org/gnss/abstracts.cfm?paperID=12345