Cross-Type Representation Learning in GNSS: A Novel Approach for Enhanced SPP Positioning through Pseudorange Residuals

Abstract In the evolving domain of satellite navigation, accurate Single Point Positioning (SPP) remains a paramount objective. After an in-depth analysis of pseudorange residuals' characteristics, this study unveils a pioneering method by harnessing cross-type representation learning within GNSS, seamlessly integrating the capabilities of natural language processing with conventional navigation methodologies. By innovatively translating diverse conditions into "literal" terms, we re-envision how navigation challenges are addressed. Our introduced C-SPP algorithm significantly surpasses established SPP methods like RTKLIB and NETDIFF, achieving a remarkable accuracy boost of 33.27% and 24.47%, respectively. Moreover, while there's a slight increment in computational time, it's impressively marginal, showcasing our method's balance between precision and efficiency. The research underscores the importance of context adaptability, emphasizing potential efficiency gains from regional applications. With a unique blend of advanced NLP and GNSS positioning methods, our study offers a refreshing lens to view and tackle pseudorange residuals, opening avenues for further research and applications.