An improved Kalman filter algorithm for tightly GNSS/INS integrated navigation system

The Kalman filter based on singular value decomposition (SVD) can sufficiently reduce the accumulation of rounding errors and is widely used in various applications with numerical calculations. However, in order to improve the filtering performance and adaptability in a tightly GNSS/INS (Global Navigation Satellite System and Inertial Navigation System) integrated navigation system, we propose an improved robust method to satisfy the requirements. To solve the issue of large fluctuations in GNSS signals faced by the conventional method that uses a fixed noise covariance, the proposed method constructs a correction variable through the innovation and the new matrix which is obtained by performing SVD on the original matrix, dynamically correcting the noise covariance and has better robustness. In addition, the derived SVD form of the information filter (IF) extends its application. The proposed method has higher positioning accuracy and can be better applied to tightly coupled GNSS/INS navigation simulations and physical experiments. The experimental results show that, compared with the traditional Kalman algorithm based on SVD, the proposed algorithm*s maximum error is reduced by 45.77%. Compared with the traditional IF algorithm, the root mean squared error of the proposed IF algorithm in the form of SVD is also reduced by 4.7%.