Deep neural network based multipath mitigation method for carrier-based differential GNSS systems

Abstract

As the range of drone applications expands, many missions are requiring a high level of navigation accuracy including drone landing, mapping, etc. Real Time Kinematic (RTK), which is one of the carrier based differential global navigation satellites systems (CD-GNSS), is getting lots of attention for drone applications. Carrier phase measurement errors are on the order of sub-centimeter level, making centimeter level navigation performance possible. However, that performance is realized when cycle integer ambiguity is successfully resolved. The commercial RTK system takes 5-10 minutes to resolve integer ambiguity without considering integrity, and the estimated integer ambiguity has large uncertainty resulting in large positioning error. To achieve a given level of integrity, RTK requires a longer filtering duration, i.e., the time necessary to resolve integer ambiguity correctly. Therefore, this study focuses on reducing filtering duration while guaranteeing integrity.

As the degree of automation increases, drone operations should be treated as safety-critical systems. For military purposes, the integrity assured carrier based landing system was developed. The Shipboard-Relative Global Positioning System (SRGPS), an architectural variant of the Joint Precision Approach and Landing System (JPALS), is a system developed to support shipboard landing with a high level of integrity (vertical protection level of 1.1 m and integrity risk of $10^{-7}$ ). Filtering duration is a function of quality of code/carrier measurements and integrity requirements. If the same level of integrity with SRGPS is applied to the carrier-phase based drone landing operation, the required filtering duration becomes about 70 minutes. This duration arises because of low code/carrier measurement quality resulting from the cheap receiver and antenna used in both reference stations and drone. In order to increase the efficiency of drone operation, we propose two methods to reduce filtering duration: 1) improve GPS measurement quality via the deep neural network based code multipath mitigation method, and 2) relax integrity requirements based on consideration of drone operation via sensitivity analysis.