ABSTRACT
Computing and updating with triangular matrices involve fewer arithmetic operations and greatly reduce the problem of round-off errors, which might cause ill-conditioning and subsequent divergence of the algorithm, especially if the filter is implemented on a finite word-length machine. The factorization algorithms are supposed to be stable and accurate compared to the information matrix and covariance algorithms. The major advantage from UDF comes from the fact that the square root–type algorithm processes square roots of the covariance matrices, and hence they essentially use half the word length normally required by the conventional Kalman filters (KFs). The information filtering (IF) is the more direct way of dealing with the target tracking and multi-sensor data fusion problems than the conventional covariance-based KF. The IF, if implemented as is, could be sensitive to computer round-off/quantization errors, which would degrade the tracking performance of the filter.
