Deep Neural Network Based Multipath Mitigation Method for Carrier Based Differential GNSS Systems

Abstract

Carrier based Differential Global Navigation Satellite System (CD-GNSS) is getting lots of attention as a promising technology for drones since it can provide centimeter-level accuracy. Unlike widely used CD-GNSS applications such as geodetic survey, CD-GNSS systems to be used for drone applications should provide a certain level of integrity along with accuracy. However, there is a critical challenge in guaranteeing the integrity of CD-GNSS systems for drone applications due to a long filter duration which is the time necessary to resolve cycle ambiguity correctly. One of the most dominant factors that limits reducing the filter duration is the code multipath error. In this response, this paper proposes a code multipath mitigation method using Deep Neural Network (DNN) for drone systems. It is well known that DNN is a powerful for nonlinear regression problems toward which multipath estimation is applicable and can be trained using a large quantity of data. The target scenario of this study is chosen as a drone flying at a sufficiently high altitude (where no multipath reflections exist from surrounding obstacles) before lowering its altitude for other phases of operation. Therefore, the dominant factor of multipath errors of the drone is the signal reflection from its body frame. Considering the fact, the input parameters were chosen for developing the DNN model as elevation, azimuth and tilt angle of antenna which characterize signals reflected by the frame and Signal to Noise Ratio (SVR) which characterizes the slight change of signals. The validity of the proposed model was investigated and its multipath mitigation performance was evaluated under the target scenario using real data. The results show that the multipath error is mitigated by about 30% in terms of the standard deviation of multipath error and the filter duration is reduced by about 66%.