Presented at ION GNSS+, September 2016
A tight coupling of GNSS and inertial measurements is needed for both accurate and reliable positioning. The use of multi-GNSS is recommended to obtain a sufficient number of visible satellites in any outdoor environment.
We perform a joint GPS/GLONASS ambiguity fixing and a tight coupling of GNSS, 3D accelerometer, 3D gyroscope, 3D magnetometer, barometer and thermometer measurements. As GLONASS uses FDMA, double difference ambiguities are no longer integer-valued. We derive a transformation for the GLONASS double difference ambiguity term, that recovers the integer property and maintains a full-rank system. The obtained transformation maps the real-valued double difference ambiguity terms into integer-valued double difference ambiguity terms and a common single difference ambiguity term, that is treated as a real-valued parameter.
Low-cost GNSS antennas cannot suppress multipath and, therefore, require an estimation of multipath errors. We provide a precise model for multipath that considers an individual amplitude, code delay, phase shift and Doppler shift for each reflected signal, and include it in our sensor fusion. The magnetometer measurements provide rough attitude information, which makes them very valuable for robust GNSS attitude ambiguity fixing.
We verified the performance of our sensor fusion in a test drive on a parking lot. The fixed phase residuals were in the order of a few centimeters for both GPS and GLONASS, which indicates a very precise position estimation. The proposed algorithms reduced the horizontal 95th-percentile error from 8.49 meters (for a standard GPS-only solution) down to 3.96 meters — a 66 percent improvement. In order to combine the GPS and VIO measurements as described in the last paragraph, the data need to be brought into the same reference frame. We develop a novel method to perform this change of reference frame. The proposed approach combines a quaternion reformulation of the problem together with a semidefinite relaxation technique.