Multi-Frequency GNSS Sensor Fusion With Quality Assessment for Automotive Applications

Abstract

The modernization of the satellite constellations of GPS and GLONASS as well as the commissioning of Galileo and BeiDou provide civil users with a significantly increased amount of navigation satellites broadcasting usable signals on multiple carrier frequencies. Simultaneously, the quality requirements for automotive localization algorithms increase, because these algorithms are no longer used exclusively for navigation. Instead, they play an important role in the development of autonomous vehicles. These increased requirements cannot be met with GNSS alone, but necessitate the fusion of data from multiple sensor types. This thesis describes the fusion of measurements from three different sensor types with an extended Kalman filter. These three sensor types are a multi-frequency GNSS receiver, an inertial measurement unit and the vehicle's built-in odometry sensors. The focus lies on the optimal processing of pseudoranges obtained from satellites of various constellations on multiple carrier frequencies. This is achieved by forming ionosphere-free linear combinations, which eliminate the largest error source of single-frequency observations. The current deployment of different active satellite generations and the varying reception conditions create the necessity to process pseudoranges or linear combinations thereof on multiple carrier frequencies simultaneously in each epoch. This requires the calibration of signal-dependent differential code biases occurring in pseudorange measurements. A suitable calibration is performed for the employed multi-frequency GNSS receiver for the GPS signals L1 C/A, L2C and L5 as well as for the Galileo signals E1, E5a and E5b. This enables the utilization of the respective pseudoranges and their linear combinations within the integration filter. In addition, a quantitative model of the measurement noise is developed and parametrized, permitting the optimal weighting of the different observations. Additional GNSS observables are time-differenced carrier phase measurements on GPS L1 C/A and Galileo E1. The vehicle's built-in odometry sensors provide the rotation rates of the four wheels and the steering wheel angle. From these quantities, the horizontal velocity vectors at the wheel contact patches are computed. During this computation, the compensation of longitudinal and lateral slip is carried out with various tire models. A major development aspect is the inclusion of correlation into the measurement noise covariance matrix. The magnitude of these correlations in lateral direction is so large that the lateral velocities of the four wheels are subsumed into a single observation per axle. After the observations from GNSS receiver and odometry sensors have been preprocessed in this way, they are fused with the measurements of a MEMS IMU in a tightly coupled integration filter. IMU error modeling is not a key aspect of this thesis and is therefore performed with conventional models. In order to assess the quality of the integrated solution supplied by the localization algorithm, performance metrics concerning accuracy and integrity are chosen. These metrics are evaluated with the help of test scenarios covering different GNSS reception conditions. The reference solution is obtained by integrating data from a ring laser gyroscope IMU and from a GNSS receiver capable of RTK positioning. The results verify that the utilization of multi-frequency observations leads to a significant accuracy improvement in all considered test scenarios. During unobstructed GNSS reception, a horizontal position error of 0.5 m or better is achieved in 95 % of epochs.